Creation and test of applied software of network of wireless sensors for agriculture
The article describes applied software of units of such complex hardware-software system, as plants` state monitoring system for application in agriculture and ecological monitoring. The mentioned system consists of data acquisition system in the form of wireless sensor network and adaptive part in...
Збережено в:
| Дата: | 2023 |
|---|---|
| Автори: | , , , , , |
| Формат: | Стаття |
| Мова: | English |
| Опубліковано: |
Інститут програмних систем НАН України
2023
|
| Теми: | |
| Онлайн доступ: | https://pp.isofts.kiev.ua/index.php/ojs1/article/view/544 |
| Теги: |
Додати тег
Немає тегів, Будьте першим, хто поставить тег для цього запису!
|
| Назва журналу: | Problems in programming |
| Завантажити файл: |  |
Репозитарії
Problems in programming| id |
pp_isofts_kiev_ua-article-544 |
|---|---|
| record_format |
ojs |
| resource_txt_mv |
ppisoftskievua/94/268021a3cbfbbf810566e06a8ea0f994.pdf |
| spelling |
pp_isofts_kiev_ua-article-5442023-06-25T08:08:05Z Creation and test of applied software of network of wireless sensors for agriculture Розробка та перевірка програмних засобів мережі бездротових сенсорів для сільського господарства Romanov, V.O. Antonova, H.V. Galelyuka, I.B. Hrusha, V.M. Kedych, A.V. Voronenko, O.V. wireless sensor network; decision-making support system; applied software; plants' state monitoring system UDC 681.3, 004.6 бездротова сенсорна мережа; система підтримки прийняття рішень; прикладний програмний засіб; система моніторингу стану рослин УДК 681.3, 004.6 The article describes applied software of units of such complex hardware-software system, as plants` state monitoring system for application in agriculture and ecological monitoring. The mentioned system consists of data acquisition system in the form of wireless sensor network and adaptive part in the form of decision-making support system. The authors described main applied software of au- tonomous nodes of wireless sensor network and implementation of some program functions of decision-making support system. Wire- less sensor network includes many autonomous wireless sensors, so the main criteria during applied software creation was assuring the energy efficiency of operation of autonomous measuring nodes and network coordinator, and correct interaction of nodes within all network. As it is very difficult to perform testing of applied software of wireless nodes individually in field conditions, it was tested the network cluster, including hardware and software as a whole, in conditions like to applied task. The main parameters, which define the correctness of applied software operation, were estimated. These parameters include, for example, time of network selforganization, distance and quality of stable communication, time of autonomous operation of wireless nodes without charging batteries and so on. To create applied software for the decision-making support system, first of all, methods of plants` state diagnosing and estimating the factors, which influence the plant state, were developed. For this, the field experiments were conducted to determine sufficient dose of herbicide application and estimate the soil moisture using the chlorophyll fluorescence induction method. For processing measured data, several methods of machine learning were used, including neural network approach. Application of machine learning methods made it possible, on the base of acquired data, to make early diagnostics of influence of stress factors on the plant even before the appearance of visual manifestations of such negative influence and determine the decrease of soil moisture through the diagnostics of plant itself, and inform the user about this.Prombles in programming 2022; 3-4: 425-436 В статті описано прикладні програмні засоби елементів такої складної апаратно-програмної системи, як система моніторингу стану рослин для застосування у сільському господарстві та екологічному моніторингу. Згадана система складається з системи збору даних у вигляді бездротової сенсорної мережі та з адаптивної частини у вигляді системи підтримки прийняття рішення. Описано основні прикладні програмні засоби вузлів бездротової сенсорної мережі та реалізацію деяких функцій системи під- тримки прийняття рішень. Бездротова сенсорна мережа складається з великої кількості автономних сенсорів, тому основним критерієм при розробці прикладних програмних засобів було гарантування енергоефективності роботи автономних вимірю- вальних вузлів і координатора мережі та їх коректної взаємодії між собою в рамках цілої мережі. Оскільки тестування роботи окремих програмних засобів елементів бездротової мережі в польових умовах провести достатньо складно, то було виконано тестування кластеру такої мережі в сукупності апаратних і програмних засобів в умовах, наближених до прикладної задачі. Оцінювалися основні параметри, які визначають коректність роботи прикладних програмних засобів, зокрема, час самоор- ганізації мережі, дальність і якість стабільного зв’язку, тривалість автономної роботи бездротових вузлів без підзарядки та ін. Для розробки прикладних програмних засобів системи підтримки прийняття рішень, в першу чергу, розроблено методики діагностики стану рослин і оцінювання чинників, які впливають на стан рослини. Для цього проведено натурні експерименти визначення достатньої дози внесення гербіциду й оцінювання вологості ґрунту за допомогою методу індукції флуоресценції хлорофілу. Для обробки отриманих результатів використано кілька методів машинного навчання, включаючи нейромережевий підхід. Застосування методів машинного навчання дало можливість на основі отриманих даних вимірювання здійснювати ран- ню діагностику впливу стресових чинників на рослину ще до появи зовнішніх візуальних проявів такого негативного впливу та визначати пониження вологості ґрунту через діагностику стану самої рослини й інформувати про це користувача.Prombles in programming 2022; 3-4: 425-436 Інститут програмних систем НАН України 2023-01-23 Article Article application/pdf https://pp.isofts.kiev.ua/index.php/ojs1/article/view/544 10.15407/pp2022.03-04.425 PROBLEMS IN PROGRAMMING; No 3-4 (2022); 425-436 ПРОБЛЕМЫ ПРОГРАММИРОВАНИЯ; No 3-4 (2022); 425-436 ПРОБЛЕМИ ПРОГРАМУВАННЯ; No 3-4 (2022); 425-436 1727-4907 10.15407/pp2022.03-04 en https://pp.isofts.kiev.ua/index.php/ojs1/article/view/544/597 Copyright (c) 2023 PROBLEMS IN PROGRAMMING |
| institution |
Problems in programming |
| baseUrl_str |
https://pp.isofts.kiev.ua/index.php/ojs1/oai |
| datestamp_date |
2023-06-25T08:08:05Z |
| collection |
OJS |
| language |
English |
| topic |
wireless sensor network decision-making support system applied software plants' state monitoring system UDC 681.3 004.6 |
| spellingShingle |
wireless sensor network decision-making support system applied software plants' state monitoring system UDC 681.3 004.6 Romanov, V.O. Antonova, H.V. Galelyuka, I.B. Hrusha, V.M. Kedych, A.V. Voronenko, O.V. Creation and test of applied software of network of wireless sensors for agriculture |
| topic_facet |
wireless sensor network decision-making support system applied software plants' state monitoring system UDC 681.3 004.6 бездротова сенсорна мережа система підтримки прийняття рішень прикладний програмний засіб система моніторингу стану рослин УДК 681.3 004.6 |
| format |
Article |
| author |
Romanov, V.O. Antonova, H.V. Galelyuka, I.B. Hrusha, V.M. Kedych, A.V. Voronenko, O.V. |
| author_facet |
Romanov, V.O. Antonova, H.V. Galelyuka, I.B. Hrusha, V.M. Kedych, A.V. Voronenko, O.V. |
| author_sort |
Romanov, V.O. |
| title |
Creation and test of applied software of network of wireless sensors for agriculture |
| title_short |
Creation and test of applied software of network of wireless sensors for agriculture |
| title_full |
Creation and test of applied software of network of wireless sensors for agriculture |
| title_fullStr |
Creation and test of applied software of network of wireless sensors for agriculture |
| title_full_unstemmed |
Creation and test of applied software of network of wireless sensors for agriculture |
| title_sort |
creation and test of applied software of network of wireless sensors for agriculture |
| title_alt |
Розробка та перевірка програмних засобів мережі бездротових сенсорів для сільського господарства |
| description |
The article describes applied software of units of such complex hardware-software system, as plants` state monitoring system for application in agriculture and ecological monitoring. The mentioned system consists of data acquisition system in the form of wireless sensor network and adaptive part in the form of decision-making support system. The authors described main applied software of au- tonomous nodes of wireless sensor network and implementation of some program functions of decision-making support system. Wire- less sensor network includes many autonomous wireless sensors, so the main criteria during applied software creation was assuring the energy efficiency of operation of autonomous measuring nodes and network coordinator, and correct interaction of nodes within all network. As it is very difficult to perform testing of applied software of wireless nodes individually in field conditions, it was tested the network cluster, including hardware and software as a whole, in conditions like to applied task. The main parameters, which define the correctness of applied software operation, were estimated. These parameters include, for example, time of network selforganization, distance and quality of stable communication, time of autonomous operation of wireless nodes without charging batteries and so on. To create applied software for the decision-making support system, first of all, methods of plants` state diagnosing and estimating the factors, which influence the plant state, were developed. For this, the field experiments were conducted to determine sufficient dose of herbicide application and estimate the soil moisture using the chlorophyll fluorescence induction method. For processing measured data, several methods of machine learning were used, including neural network approach. Application of machine learning methods made it possible, on the base of acquired data, to make early diagnostics of influence of stress factors on the plant even before the appearance of visual manifestations of such negative influence and determine the decrease of soil moisture through the diagnostics of plant itself, and inform the user about this.Prombles in programming 2022; 3-4: 425-436 |
| publisher |
Інститут програмних систем НАН України |
| publishDate |
2023 |
| url |
https://pp.isofts.kiev.ua/index.php/ojs1/article/view/544 |
| work_keys_str_mv |
AT romanovvo creationandtestofappliedsoftwareofnetworkofwirelesssensorsforagriculture AT antonovahv creationandtestofappliedsoftwareofnetworkofwirelesssensorsforagriculture AT galelyukaib creationandtestofappliedsoftwareofnetworkofwirelesssensorsforagriculture AT hrushavm creationandtestofappliedsoftwareofnetworkofwirelesssensorsforagriculture AT kedychav creationandtestofappliedsoftwareofnetworkofwirelesssensorsforagriculture AT voronenkoov creationandtestofappliedsoftwareofnetworkofwirelesssensorsforagriculture AT romanovvo rozrobkataperevírkaprogramnihzasobívmerežíbezdrotovihsensorívdlâsílʹsʹkogogospodarstva AT antonovahv rozrobkataperevírkaprogramnihzasobívmerežíbezdrotovihsensorívdlâsílʹsʹkogogospodarstva AT galelyukaib rozrobkataperevírkaprogramnihzasobívmerežíbezdrotovihsensorívdlâsílʹsʹkogogospodarstva AT hrushavm rozrobkataperevírkaprogramnihzasobívmerežíbezdrotovihsensorívdlâsílʹsʹkogogospodarstva AT kedychav rozrobkataperevírkaprogramnihzasobívmerežíbezdrotovihsensorívdlâsílʹsʹkogogospodarstva AT voronenkoov rozrobkataperevírkaprogramnihzasobívmerežíbezdrotovihsensorívdlâsílʹsʹkogogospodarstva |
| first_indexed |
2024-09-12T19:29:52Z |
| last_indexed |
2024-09-12T19:29:52Z |
| _version_ |
1815407583810289664 |
| fulltext |
425
Прикладне програмне забезпечення
УДК 681.3, 004.6 https://doi.org/10.15407/pp2022.03-04.425
CREATION AND TEST OF APPLIED
SOFTWARE OF NETWORK OF WIRELESS SENSORS
FOR AGRICULTURE
Volodymyr Romanov, Hanna Antonova, Igor Galelyuka,
Volodymyr Hrusha, Anna Kedych, Oleksandr Voronenko
В статті описано прикладні програмні засоби елементів такої складної апаратно-програмної системи, як система моніторингу
стану рослин для застосування у сільському господарстві та екологічному моніторингу. Згадана система складається з системи
збору даних у вигляді бездротової сенсорної мережі та з адаптивної частини у вигляді системи підтримки прийняття рішення.
Описано основні прикладні програмні засоби вузлів бездротової сенсорної мережі та реалізацію деяких функцій системи під-
тримки прийняття рішень. Бездротова сенсорна мережа складається з великої кількості автономних сенсорів, тому основним
критерієм при розробці прикладних програмних засобів було гарантування енергоефективності роботи автономних вимірю-
вальних вузлів і координатора мережі та їх коректної взаємодії між собою в рамках цілої мережі. Оскільки тестування роботи
окремих програмних засобів елементів бездротової мережі в польових умовах провести достатньо складно, то було виконано
тестування кластеру такої мережі в сукупності апаратних і програмних засобів в умовах, наближених до прикладної задачі.
Оцінювалися основні параметри, які визначають коректність роботи прикладних програмних засобів, зокрема, час самоор-
ганізації мережі, дальність і якість стабільного зв’язку, тривалість автономної роботи бездротових вузлів без підзарядки та
ін. Для розробки прикладних програмних засобів системи підтримки прийняття рішень, в першу чергу, розроблено методики
діагностики стану рослин і оцінювання чинників, які впливають на стан рослини. Для цього проведено натурні експерименти
визначення достатньої дози внесення гербіциду й оцінювання вологості ґрунту за допомогою методу індукції флуоресценції
хлорофілу. Для обробки отриманих результатів використано кілька методів машинного навчання, включаючи нейромережевий
підхід. Застосування методів машинного навчання дало можливість на основі отриманих даних вимірювання здійснювати ран-
ню діагностику впливу стресових чинників на рослину ще до появи зовнішніх візуальних проявів такого негативного впливу
та визначати пониження вологості ґрунту через діагностику стану самої рослини й інформувати про це користувача.
Ключові слова: бездротова сенсорна мережа, система підтримки прийняття рішень, прикладний програмний засіб, система мо-
ніторингу стану рослин.
The article describes applied software of units of such complex hardware-software system, as plants` state monitoring system for ap-
plication in agriculture and ecological monitoring. The mentioned system consists of data acquisition system in the form of wireless
sensor network and adaptive part in the form of decision-making support system. The authors described main applied software of au-
tonomous nodes of wireless sensor network and implementation of some program functions of decision-making support system. Wire-
less sensor network includes many autonomous wireless sensors, so the main criteria during applied software creation was assuring
the energy efficiency of operation of autonomous measuring nodes and network coordinator, and correct interaction of nodes within all
network. As it is very difficult to perform testing of applied software of wireless nodes individually in field conditions, it was tested the
network cluster, including hardware and software as a whole, in conditions like to applied task. The main parameters, which define the
correctness of applied software operation, were estimated. These parameters include, for example, time of network self-organization,
distance and quality of stable communication, time of autonomous operation of wireless nodes without charging batteries and so on.
To create applied software for the decision-making support system, first of all, methods of plants` state diagnosing and estimating the
factors, which influence the plant state, were developed. For this, the field experiments were conducted to determine sufficient dose
of herbicide application and estimate the soil moisture using the chlorophyll fluorescence induction method. For processing measured
data, several methods of machine learning were used, including neural network approach. Application of machine learning methods
made it possible, on the base of acquired data, to make early diagnostics of influence of stress factors on the plant even before the
appearance of visual manifestations of such negative influence and determine the decrease of soil moisture through the diagnostics of
plant itself, and inform the user about this.
Keywords: wireless sensor network, decision-making support system, applied software, plants` state monitoring system.
Introduction
The speed and the accuracy of decision making in ecological monitoring, precision or digital agriculture are de-
termined usually by amount, quality and frequency of acquiring of measuring data from monitored territories or objects
and by the speed of data processing. A set of mobile and traditional data acquisition technologies, databases and software
for preliminary data processing can be called as a base system for decision-making support. The next unit, main purpose
of which is generation and optimization of the made managerial decision and further actions, can be called as adaptive
unit. Data, which are acquired directly from monitored territories or objects, include meteorological data, information
about soils` state and, no less important, information about plants` state.
Therefore, the acquisition of operative and objective information about the state of plant cover in many cases
is an important factor, which causes the further strategy of keeping agricultural lands, greenery of megalopolises, na-
tional plant reserves, woodlands and parklands and making the proper optimal decision. Certainly, it would be ideal
to acquire information about the impairment or improvement of state of plants beforehand, but not long time after the
event. It will allow to save plants from possible losses, reduce the costs for their protection, and decrease the negative
impact of stress factors of natural or anthropogenic origin on plants.
It should be noted, that mobile and traditional data acquisition system play an important role as in precision and digi-
© В.О. Романов, Г.В. Антонова, І.Б. Галелюка, В.М. Груша, А.В. Кедич, О.В. Вороненко, 2022
ISSN 1727-4907. Проблеми програмування. 2022. № 3-4. Спеціальний випуск
426
Прикладне програмне забезпечення
tal agriculture, so in ecological monitoring and environmental protection. Currently, during the fast development of wireless
technologies and Internet of Things, it is reasonable as mobile data acquisition system to use the wireless sensor networks,
the devices of Internet of Things or Industrial Internet of Things in combination with wireless technologies of long distance
communication.
Now many projects, concerning the application of wireless sensor networks in the various areas of human
activity, have already been completed and implemented or are under the researches and development. There are some
projects and system, in which wireless sensor network are used for data acquisition in digital agriculture or ecological
monitoring. Swiss commercial system “Predicting Diseases of Vine (PreDiVine)” [1] is intended to fight against pests
of vine. Wireless sensors are allocated on buds and stems of the vine for measuring air temperature. Sensors transmit
the data via wireless channel to computer device, which models the evolution of pest on the base of temperature data.
Hungarian company “AgroSense” [2] developed wireless sensor network for agricultural sector. Network collects data
of soil temperature, and humidity, leaf wetness and thus controls the watering. In work [3] the wireless sensor network
are used to collect environments parameters to protect plants from frosts. Paper [4] describes a successful solution for
ecological monitoring, where a wireless sensor network is implemented and used to monitor state of plants and trees,
which can operate as in field conditions, so in indoor conditions. It should be noted that there are many similar projects
with almost similar functionality and use hardware-software tools.
We want to point, that in all mentioned projects, technologies and examined papers proposed systems for mea-
suring different parameters of soil, plants and environment including temperature, humidity, biochemical parameters
of plants, even sound range etc. But any system does not measure parameters of photosynthesis – the main process of
plant vital activity. If it is necessary to estimate plants state and influence of stress factors of natural or anthropogenic
origin on them, the most effective and informative way for accurate estimating of plant state is to measure the photo-
synthesis parameters in plants [5].
Data acquisition system implementation
Modern achievements in microelectronics, biosensor technologies, and information and communication tech-
nologies make it possible to develop and create wireless sensor networks for express-diagnostics of plants` state on
large territories using an effect of chlorophyll fluorescence induction. Moreover, the integration the miniature sensors
into wireless sensor networks to measure additional parameters of state of air and soil make it possible to diagnose not
only the general state of plants, but also such parameters, as pesticides content in soil, water and plants, level of soil
corrosion, air pollution and so on.
The elements of proposed data acquisition systems are wireless sensor networks or their clusters. The cover
area of such network can varies from several square metres to several square kilometres due to possibility to transfer
measured data from one network node to another. The main advantage of wireless sensor network is ability to real-time
monitor the state of agricultural plants or environmental parameters on large territories. Fig. 1 shows the structure of
proposed data acquisition system for agriculture or ecological monitoring.Прикладне програмне забезпечення
Fig. 1. Structure of data acquisition system for agriculture and ecological monitoring
The feature of proposed system is that the individual wireless sensor network can cover the territories of differ-
ent areas, so they have to be very well scalable. Scalability causes the presence some number of signal repeaters and
proper network self-organization algorithms in the network. In addition, the networks can be located on different dis-
tances from server, as well as each from other. Such preconditions of deployment of networks and system in whole
make it reasonable to use a several protocols of different effective distance to transfer data. To organize the communica-
tion between nodes of separate networks it is reasonable to use industrial wireless communication protocols or technol-
ogies of Industrial Internet of Things, for example, ZigBee or Bluetooth Mesh. To organize the communication between
the separate networks and server, it is advisable, if it is necessary, to use technologies of long-distance communication,
for example, 4G LTE.
Wireless microcontroller of nRF52840 type were used to create wireless measuring nodes, signal repeaters, con-
trol nodes and network coordinators. The generalized structural diagram of the wireless communication module on the
base of system-on-chip nRF52840 with main interfaces, implemented for our applied task, is shown on Fig. 2.
Fig. 2. The generalized structural diagram of the wireless communication module
Connection of different interfaces make it possible to create different network nodes. The measuring node pro-
vides the connection of wired channel UART and wireless channel ZigBee or Bluetooth. The network coordinator pro-
vides the connection of wireless channel ZigBee or Bluetooth and additional units, including display, memory and con-
trol elements. In addition, in some applied tasks the coordinator may need the connection via wired channel with PC.
The next network unit is control node, which gives possibility to connect measuring nodes directly to a PC. The control
node provides the connection of wireless channel ZigBee or Bluetooth and wired channel USBD for connecting to a PC.
Fig. 1. Structure of data acquisition system for agriculture and ecological monitoring
427
Прикладне програмне забезпечення
Acquiring data about state of monitored objects is made with wireless sensors, which are combined into
wireless sensor network. In this case, the defined territory can be covered as by separate wireless network, so
the cluster of such network. Of course, a certain number of measuring nodes can be equipped with wired data
transfer channels and connected to single wireless node, which will aggregate the acquired measured data and
transfer them to the next network node. There are two types of wireless measuring nodes. The first ones are in-
tended to measure the state of alive plant by measuring the chlorophyll fluorescence induction, the second ones
– to measure the parameters of soil and air. Measured data are transmitted through signal repeaters to the network
coordinator, which aggregates measured data from different measuring nodes and send them to the server. The
special software module for preliminary data processing is on the server side. The main purpose of this module is
preparing data for making managerial decision or measured data visualization.
The set of measuring tools and server with software module for preliminary data processing and data visualiza-
tion is the basic part of system, which is intended to acquaint users with state of monitored territories or objects in real
time mode. Software unit of generation and optimization of managerial decision refers to adaptive part of the system,
the main purpose of which is to help user in keeping and caring the monitored territories or objects and making the
managerial decision to minimize the impact of negative factors of natural or anthropogenic origin.
Depending of applied task, the user can deploy basic configuration of data acquisition system or full system
with adaptive part. At basic configuration, the user receives only information about state of monitored territory or
objects in visual mode and then the user itself makes the managerial decision about further actions. When deploying
the full system with adaptive part, the user receives not only the information about state of monitored territories and
objects, but also gets managerial decision or recommendations about further actions, produced by the adaptive part of
system.
The feature of proposed system is that the individual wireless sensor network can cover the territories of
different areas, so they have to be very well scalable. Scalability causes the presence some number of signal re-
peaters and proper network self-organization algorithms in the network. In addition, the networks can be located
on different distances from server, as well as each from other. Such preconditions of deployment of networks and
system in whole make it reasonable to use a several protocols of different effective distance to transfer data. To
organize the communication between nodes of separate networks it is reasonable to use industrial wireless com-
munication protocols or technologies of Industrial Internet of Things, for example, ZigBee or Bluetooth Mesh.
To organize the communication between the separate networks and server, it is advisable, if it is necessary, to use
technologies of long-distance communication, for example, 4G LTE.
Wireless microcontroller of nRF52840 type were used to create wireless measuring nodes, signal repeaters,
control nodes and network coordinators. The generalized structural diagram of the wireless communication module
on the base of system-on-chip nRF52840 with main interfaces, implemented for our applied task, is shown on Fig. 2.
Прикладне програмне забезпечення
Fig. 1. Structure of data acquisition system for agriculture and ecological monitoring
The feature of proposed system is that the individual wireless sensor network can cover the territories of differ-
ent areas, so they have to be very well scalable. Scalability causes the presence some number of signal repeaters and
proper network self-organization algorithms in the network. In addition, the networks can be located on different dis-
tances from server, as well as each from other. Such preconditions of deployment of networks and system in whole
make it reasonable to use a several protocols of different effective distance to transfer data. To organize the communica-
tion between nodes of separate networks it is reasonable to use industrial wireless communication protocols or technol-
ogies of Industrial Internet of Things, for example, ZigBee or Bluetooth Mesh. To organize the communication between
the separate networks and server, it is advisable, if it is necessary, to use technologies of long-distance communication,
for example, 4G LTE.
Wireless microcontroller of nRF52840 type were used to create wireless measuring nodes, signal repeaters, con-
trol nodes and network coordinators. The generalized structural diagram of the wireless communication module on the
base of system-on-chip nRF52840 with main interfaces, implemented for our applied task, is shown on Fig. 2.
Fig. 2. The generalized structural diagram of the wireless communication module
Connection of different interfaces make it possible to create different network nodes. The measuring node pro-
vides the connection of wired channel UART and wireless channel ZigBee or Bluetooth. The network coordinator pro-
vides the connection of wireless channel ZigBee or Bluetooth and additional units, including display, memory and con-
trol elements. In addition, in some applied tasks the coordinator may need the connection via wired channel with PC.
The next network unit is control node, which gives possibility to connect measuring nodes directly to a PC. The control
node provides the connection of wireless channel ZigBee or Bluetooth and wired channel USBD for connecting to a PC.
Fig. 2. The generalized structural diagram of the wireless communication module
Connection of different interfaces make it possible to create different network nodes. The measuring node pro-
vides the connection of wired channel UART and wireless channel ZigBee or Bluetooth. The network coordinator pro-
vides the connection of wireless channel ZigBee or Bluetooth and additional units, including display, memory and control
elements. In addition, in some applied tasks the coordinator may need the connection via wired channel with PC. The
next network unit is control node, which gives possibility to connect measuring nodes directly to a PC. The control node
provides the connection of wireless channel ZigBee or Bluetooth and wired channel USBD for connecting to a PC.
The proposed network elements, – measuring node, control node and network coordinator, – make it possible
to build several network topologies on the principle of pico-network. It should be noted, that for some applied tasks the
network coordinator can be replaced by a smartphone or tablet computer with special applied software, which fully imple-
ments the functions of network coordinator. The control node can operate only in combination with a PC or, as an option,
a laptop. Therefore, the main purpose of control node is to connect ZigBee- or Bluetooth-network to PC or laptop.
428
Прикладне програмне забезпечення
Implementation of applied software of network nodes
Mentioned above network nodes usually can be adapted to different applied tasks through small update of applied
software and reconfiguration of interfaces of wireless nodes. At the same time, the hardware does not need any changes or
upgrades, with the exception of rechargeable batteries or memory modules. Each of proposed wireless nodes has a certain
functional purpose, for the implementation of functions of which the appropriate applied software have been created.
The network coordinator is intended for network organization, connection of measuring nodes, visualization
data about network, measuring nodes or measured data, and, if it is necessary, connection to a PC for data transferring.
The network coordinator works independently and the presence of PC is not mandatory. The main functions of network
coordinator, which are implemented through appropriate applied software or program services, are the next:
– organization of wireless connections with individual measuring nodes;
– formation of wireless sensor network and control of its topology;
– setting and configuration of network parameters, setting and maintenance of network politic;
– setting the criteria of connections of wireless measuring nodes;
– control the measurements in whole network or individual measuring nodes;
– storage of the measured data from all measuring nodes;
– measured data processing and aggregation;
– visualization of information about whole network or individual measuring nodes;
– visualization of measured data;
– establishing a connection with PC to transfer data or receive settings.
The main purpose of wireless measuring node is to measure by the sensor, connected via UART interface, in our
case. The main functions of measuring node, which are implemented through appropriate applied software or program
services, are the next:
– connection to random or predetermined wireless sensor network;
– conducting measurements according to previously set program or mode;
– preliminary processing of measured data;
– storage of certain number of measurements;
– data transferring to network coordinator, mobile device or via control node to a PC.
Control node is intended to organize the wireless network, connect the measuring nodes and establish the con-
nection of wireless network with PC or laptop. The main functions of control node, which are implemented through ap-
propriate applied software or program services, are the next:
– organization of wireless connections with individual measuring nodes;
– formation of wireless sensor network and control of its topology;
– setting and configuration of network parameters, setting and maintenance of network politic;
– setting the criteria of connections of wireless measuring nodes;
– establishing the transit data channel to transfer data from network nodes to a PC;
– transfer of control commands from PC to all wireless measuring nodes or selected one;
– transfer of measured data from measuring nodes to a PC.
Different applied software has been created for network coordinator, measuring node and control node, so their
operating algorithms slightly differ. After switching on, the network coordinator reads the parameters of network and
interfaces from internal memory or these parameters can be entered by user via proper control units. Then, timers and
counters, internal interfaces (SPI, UART, USBD, etc.), control units (buttons, swithers etc.) and display elements (indi-
cators, LCD display etc.) are initialized and started according to received parameters. Then the selected power profile
is configured and started. The last preparatory stage is initializing and starting the ZigBee or Bluetooth protocol stack,
proper attendant services and filters of possible connections. Then it is started the process of scanning the radio-air and
searching the potential nodes to connect according to predefined filters. In parallel, the working mode of the coordinator
starts, in which the program module for analysis and processing the events and interrupts also begins to work. The mod-
ule for scanning and searching the potential nodes determines whether the maximum number of connections has been
reached and, if not, it connects the found node to the network. Such operation is repeated until the maximum number of
connections is reached. In this case, the module for scanning and searching the potential nodes stops its operating. If dur-
ing the working cycle any node disconnects from the network, then the module for scanning and searching the potential
nodes starts operating to connect new found nodes to network until the maximum number of connections is reached. The
generalized structure of applied software of the network coordinator is shown on the Fig. 3.
During the work cycle of the network coordinator, the module for analysis and processing of events and
interruptions is constantly running. As soon as event or interruption is detected, the event handler is run to determine
the type of event or interruption. If the event or interruption relates to stopping operation of coordinator, disconnect-
ing of connected nodes, data processing or data transferring, then the proper software module of handler is started.
If the command to stop the coordinator operation is received, then the coordinator stops its software and hardware
modules in regular mode and, only after that, the power of coordinator itself is switched off. If the event of node
disconnecting is detected, then the module for scanning and searching the potential nodes starts operating and con-
nects the new potential nodes. If the event of input data presence is detected, then the analysing packet module is
started to analyse the format of message and determine the content of data packet. If data packet includes measured
data, such data is preliminary processed and stored in internal memory. If data packet includes command or service
data, then the analysing packet module launches proper software module.
429
Прикладне програмне забезпеченняПрикладне програмне забезпечення
Fig. 3. Structure of applied software of network coordinator
During the work cycle of the network coordinator, the module for analysis and processing of events and interrup-
tions is constantly running. As soon as event or interruption is detected, the event handler is run to determine the type of
event or interruption. If the event or interruption relates to stopping operation of coordinator, disconnecting of connect-
ed nodes, data processing or data transferring, then the proper software module of handler is started. If the command to
stop the coordinator operation is received, then the coordinator stops its software and hardware modules in regular
mode and, only after that, the power of coordinator itself is switched off. If the event of node disconnecting is detected,
then the module for scanning and searching the potential nodes starts operating and connects the new potential nodes. If
the event of input data presence is detected, then the analysing packet module is started to analyse the format of mes-
sage and determine the content of data packet. If data packet includes measured data, such data is preliminary processed
and stored in internal memory. If data packet includes command or service data, then the analysing packet module
launches proper software module.
After switching on, the measuring node reads the parameters of connection to network and operational mode
from internal memory. Then, timers and counters, internal interfaces (UART, etc.), control units (buttons, swithers etc.)
and display elements (indicators, etc.) are initialized and started according to read parameters. Then the selected power
profile is configured and started. The last preparatory stage is initializing and starting the ZigBee or Bluetooth protocol
stack and proper attendant services. Then it is started the searching of the main network device to connect accordingly
to the read connection parameters. After connection to the main network device, it is started the working cycle of meas-
uring node. The generalized structure of applied software of the measuring node is shown on the Fig. 4.
During the work cycle of the measuring node, the module for analysis and processing of events and interruptions
is constantly running. As soon as event or interruption is detected, the event handler is run to determine the type of
event or interruption. If the event or interruption relates to stopping operation of measuring node, starting the measure-
ment or input data processing, then the proper software module of handler is started. If the command to stop the meas-
uring node operation is received, then the measuring node stops its software and hardware modules in regular mode
and, only after that, the power of node itself is switched off. If the command to start measurement is received, then the
proper software module conducts measurement accordingly to selected mode, and measured data are stored in internal
memory or transferred to main network device. If the event of input data presence is detected, then the analysing packet
module is started to analyse the format of message and determine the content of data packet. If data packet includes
command or service data, then the analysing packet module launches one of software modules accordingly to type of
command or service data. If data packet includes measured data from sensor interface, then it is started the software
module for processing the measured data, which makes decision to store them in internal memory or transfer to main
network device accordingly to actual operational parameters.
Fig. 3. Structure of applied software of network coordinator
After switching on, the measuring node reads the parameters of connection to network and operational mode
from internal memory. Then, timers and counters, internal interfaces (UART, etc.), control units (buttons, swithers
etc.) and display elements (indicators, etc.) are initialized and started according to read parameters. Then the selected
power profile is configured and started. The last preparatory stage is initializing and starting the ZigBee or Bluetooth
protocol stack and proper attendant services. Then it is started the searching of the main network device to connect
accordingly to the read connection parameters. After connection to the main network device, it is started the working
cycle of measuring node. The generalized structure of applied software of the measuring node is shown on the Fig. 4.
Прикладне програмне забезпечення
Fig. 4. Structure of applied software of measuring node
Interaction of applied software of network nodes
Applied software was developed for every network node, which interact with each other via special software in-
terfaces and services during the network operation. The scheme of interaction of applied software of wireless nodes in
the network on the base of network coordinator is shown on Fig. 5, on the base of control node is shown on Fig. 6.
Fig. 5. The scheme of interaction of applied software of wireless nodes on the base of network coordinator
When network is organized on the base of network coordinator, the presence of PC is optional. Applied software
of network coordinator plays the main role in network organizing, maintaining its operation and controlling both the
network in whole and individual wireless nodes. Data processing and storing also completely depend on applied soft-
ware of network coordinator. PC in such scheme serves only as additional tool for storage and visualization of measured
data.
When network is organized on the base of control node, the applied software of network coordinator only organ-
izes the network and maintains its operation. In such scheme, the applied software of control node plays the main role
Fig. 4. Structure of applied software of measuring node
430
Прикладне програмне забезпечення
During the work cycle of the measuring node, the module for analysis and processing of events and
interruptions is constantly running. As soon as event or interruption is detected, the event handler is run to de-
termine the type of event or interruption. If the event or interruption relates to stopping operation of measuring
node, starting the measurement or input data processing, then the proper software module of handler is started.
If the command to stop the measuring node operation is received, then the measuring node stops its software
and hardware modules in regular mode and, only after that, the power of node itself is switched off. If the com-
mand to start measurement is received, then the proper software module conducts measurement accordingly to
selected mode, and measured data are stored in internal memory or transferred to main network device. If the
event of input data presence is detected, then the analysing packet module is started to analyse the format of
message and determine the content of data packet. If data packet includes command or service data, then the
analysing packet module launches one of software modules accordingly to type of command or service data. If
data packet includes measured data from sensor interface, then it is started the software module for processing
the measured data, which makes decision to store them in internal memory or transfer to main network device
accordingly to actual operational parameters.
Interaction of applied software of network nodes
Applied software was developed for every network node, which interact with each other via special
software interfaces and services during the network operation. The scheme of interaction of applied software
of wireless nodes in the network on the base of network coordinator is shown on Fig. 5, on the base of control
node is shown on Fig. 6.
Прикладне програмне забезпечення
Fig. 4. Structure of applied software of measuring node
Interaction of applied software of network nodes
Applied software was developed for every network node, which interact with each other via special software in-
terfaces and services during the network operation. The scheme of interaction of applied software of wireless nodes in
the network on the base of network coordinator is shown on Fig. 5, on the base of control node is shown on Fig. 6.
Fig. 5. The scheme of interaction of applied software of wireless nodes on the base of network coordinator
When network is organized on the base of network coordinator, the presence of PC is optional. Applied software
of network coordinator plays the main role in network organizing, maintaining its operation and controlling both the
network in whole and individual wireless nodes. Data processing and storing also completely depend on applied soft-
ware of network coordinator. PC in such scheme serves only as additional tool for storage and visualization of measured
data.
When network is organized on the base of control node, the applied software of network coordinator only organ-
izes the network and maintains its operation. In such scheme, the applied software of control node plays the main role
Fig. 5. The scheme of interaction of applied software of wireless nodes
on the base of network coordinator
When network is organized on the base of network coordinator, the presence of PC is optional. Applied
software of network coordinator plays the main role in network organizing, maintaining its operation and con-
trolling both the network in whole and individual wireless nodes. Data processing and storing also completely
depend on applied software of network coordinator. PC in such scheme serves only as additional tool for storage
and visualization of measured data.
When network is organized on the base of control node, the applied software of network coordinator only or-
ganizes the network and maintains its operation. In such scheme, the applied software of control node plays the main
role and controls all the measuring nodes, gives command to conduct measurements, processes, stores and visualizes
the measured data. At the same time, the applied software of network coordinator switches to “tunnel” mode, at which
it only transfers commands directly from PC to nodes and measured data in reverse direction.
431
Прикладне програмне забезпечення
Прикладне програмне забезпечення
and controls all the measuring nodes, gives command to conduct measurements, processes, stores and visualizes the
measured data. At the same time, the applied software of network coordinator switches to “tunnel” mode, at which it
only transfers commands directly from PC to nodes and measured data in reverse direction.
Fig. 6. The scheme of interaction of applied software of wireless nodes on the base of control node
Implementation of functions of applied software for generation and optimization of
managerial decisions
As mentioned above, the user can deploy the system for agriculture monitoring both with adaptive module and
without it. In the first case, the user will receive some managerial decisions depending on acquired measured data. In
the second case, the user will need to make some decisions itself on the base of information about plants` state on cer-
tain territory or territories, presented in text, table or graphic form.
In order to make managerial decisions, a number of applied methods of diagnostics of plant state already have
been developed in the world with using the machine learning methods to process the measurements of chlorophyll fluo-
rescence induction. Among them, it is possible to note out the forecasting state of ornamental plants depending the envi-
ronmental conditions and changes in the chlorophyll fluorescence induction [6], taxonomic distinction of plants by
chlorophyll fluorescence curves [7], determination of water deficit [8, 9], determination of stress source [10], determi-
nation of plant maturity [11] and etc.
For adaptive part, that is decision-making support system, the authors have developed the applied method for de-
termination of sufficient dose of herbicide application using the chlorophyll fluorescence induction. For the purpose the
proper experiment was conducted, in which the plant objects were divided into three groups (control – without herbi-
cide application, and other two groups with application of different doses of herbicide). The collected data were used to
study the possibilities of different machine learning algorithms to classify plants into proper groups on different days of
experiment.
The neural network approach used the two-layer neural network with sigmoid function of activation (hyperbolic
tangent) and normalized exponential (softmax) function on the output layer of neurons. The used neural network
showed, that already on the seventh day after herbicide application of plants, it could learn to recognize the plants of
different groups. The normalization of the curve made it possible to improve the result, what eliminated some differ-
ences between curves, measured by different sensors. The best result was shown by the minimax normalization within
range [–1, 1] (1) and the centralization of the value around the average (z–score) (2):
12~
,0,
,0 −
−
−
=
kkm
kik
ik FF
Fx
x , (1)
kx
kik
ik
xxx
−
=~ , (2)
Fig. 6. The scheme of interaction of applied software of wireless nodes on the base of control node
Implementation of functions of applied software for generation and optimization of
managerial decisions
As mentioned above, the user can deploy the system for agriculture monitoring both with adaptive module
and without it. In the first case, the user will receive some managerial decisions depending on acquired measured data.
In the second case, the user will need to make some decisions itself on the base of information about plants` state on
certain territory or territories, presented in text, table or graphic form.
In order to make managerial decisions, a number of applied methods of diagnostics of plant state already have been
developed in the world with using the machine learning methods to process the measurements of chlorophyll fluorescence
induction. Among them, it is possible to note out the forecasting state of ornamental plants depending the environmental condi-
tions and changes in the chlorophyll fluorescence induction [6], taxonomic distinction of plants by chlorophyll fluorescence
curves [7], determination of water deficit [8, 9], determination of stress source [10], determination of plant maturity [11] and etc.
For adaptive part, that is decision-making support system, the authors have developed the applied method
for determination of sufficient dose of herbicide application using the chlorophyll fluorescence induction. For the
purpose the proper experiment was conducted, in which the plant objects were divided into three groups (control –
without herbicide application, and other two groups with application of different doses of herbicide). The collected
data were used to study the possibilities of different machine learning algorithms to classify plants into proper
groups on different days of experiment.
The neural network approach used the two-layer neural network with sigmoid function of activation (hyper-
bolic tangent) and normalized exponential (softmax) function on the output layer of neurons. The used neural network
showed, that already on the seventh day after herbicide application of plants, it could learn to recognize the plants of
different groups. The normalization of the curve made it possible to improve the result, what eliminated some differ-
ences between curves, measured by different sensors. The best result was shown by the minimax normalization within
range [–1, 1] (1) and the centralization of the value around the average (z–score) (2):
Прикладне програмне забезпечення
and controls all the measuring nodes, gives command to conduct measurements, processes, stores and visualizes the
measured data. At the same time, the applied software of network coordinator switches to “tunnel” mode, at which it
only transfers commands directly from PC to nodes and measured data in reverse direction.
Fig. 6. The scheme of interaction of applied software of wireless nodes on the base of control node
Implementation of functions of applied software for generation and optimization of
managerial decisions
As mentioned above, the user can deploy the system for agriculture monitoring both with adaptive module and
without it. In the first case, the user will receive some managerial decisions depending on acquired measured data. In
the second case, the user will need to make some decisions itself on the base of information about plants` state on cer-
tain territory or territories, presented in text, table or graphic form.
In order to make managerial decisions, a number of applied methods of diagnostics of plant state already have
been developed in the world with using the machine learning methods to process the measurements of chlorophyll fluo-
rescence induction. Among them, it is possible to note out the forecasting state of ornamental plants depending the envi-
ronmental conditions and changes in the chlorophyll fluorescence induction [6], taxonomic distinction of plants by
chlorophyll fluorescence curves [7], determination of water deficit [8, 9], determination of stress source [10], determi-
nation of plant maturity [11] and etc.
For adaptive part, that is decision-making support system, the authors have developed the applied method for de-
termination of sufficient dose of herbicide application using the chlorophyll fluorescence induction. For the purpose the
proper experiment was conducted, in which the plant objects were divided into three groups (control – without herbi-
cide application, and other two groups with application of different doses of herbicide). The collected data were used to
study the possibilities of different machine learning algorithms to classify plants into proper groups on different days of
experiment.
The neural network approach used the two-layer neural network with sigmoid function of activation (hyperbolic
tangent) and normalized exponential (softmax) function on the output layer of neurons. The used neural network
showed, that already on the seventh day after herbicide application of plants, it could learn to recognize the plants of
different groups. The normalization of the curve made it possible to improve the result, what eliminated some differ-
ences between curves, measured by different sensors. The best result was shown by the minimax normalization within
range [–1, 1] (1) and the centralization of the value around the average (z–score) (2):
12~
,0,
,0 −
−
−
=
kkm
kik
ik FF
Fx
x , (1)
kx
kik
ik
xxx
−
=~ , (2)
(1)
Прикладне програмне забезпечення
and controls all the measuring nodes, gives command to conduct measurements, processes, stores and visualizes the
measured data. At the same time, the applied software of network coordinator switches to “tunnel” mode, at which it
only transfers commands directly from PC to nodes and measured data in reverse direction.
Fig. 6. The scheme of interaction of applied software of wireless nodes on the base of control node
Implementation of functions of applied software for generation and optimization of
managerial decisions
As mentioned above, the user can deploy the system for agriculture monitoring both with adaptive module and
without it. In the first case, the user will receive some managerial decisions depending on acquired measured data. In
the second case, the user will need to make some decisions itself on the base of information about plants` state on cer-
tain territory or territories, presented in text, table or graphic form.
In order to make managerial decisions, a number of applied methods of diagnostics of plant state already have
been developed in the world with using the machine learning methods to process the measurements of chlorophyll fluo-
rescence induction. Among them, it is possible to note out the forecasting state of ornamental plants depending the envi-
ronmental conditions and changes in the chlorophyll fluorescence induction [6], taxonomic distinction of plants by
chlorophyll fluorescence curves [7], determination of water deficit [8, 9], determination of stress source [10], determi-
nation of plant maturity [11] and etc.
For adaptive part, that is decision-making support system, the authors have developed the applied method for de-
termination of sufficient dose of herbicide application using the chlorophyll fluorescence induction. For the purpose the
proper experiment was conducted, in which the plant objects were divided into three groups (control – without herbi-
cide application, and other two groups with application of different doses of herbicide). The collected data were used to
study the possibilities of different machine learning algorithms to classify plants into proper groups on different days of
experiment.
The neural network approach used the two-layer neural network with sigmoid function of activation (hyperbolic
tangent) and normalized exponential (softmax) function on the output layer of neurons. The used neural network
showed, that already on the seventh day after herbicide application of plants, it could learn to recognize the plants of
different groups. The normalization of the curve made it possible to improve the result, what eliminated some differ-
ences between curves, measured by different sensors. The best result was shown by the minimax normalization within
range [–1, 1] (1) and the centralization of the value around the average (z–score) (2):
12~
,0,
,0 −
−
−
=
kkm
kik
ik FF
Fx
x , (1)
kx
kik
ik
xxx
−
=~ , (2)
(2)
where
Прикладне програмне забезпечення
where ikх – matrix element in i row, k column; ikх~ – normalized fluorescence value in i row, k column; kх –
average value of k column;
kх – standard deviation of chlorophyll fluorescence induction values in k column;. kmF , –
maximum value in k column, kF ,0 – minimum fluorescence value in k column.
The application of these methods of normalization made it possible to improve the average accuracy of recogni-
tion by neural network of groups of plants, treated with herbicide, on the seventh day after herbicide application from
66,7 % to 80,9 %. Neural networks provide an early determination of influence of stress factor on plant state even be-
fore appearance of visual signs on the plant. They showed good results on the seventh day after herbicide application,
while the visual signs of the herbicide influence only appear on the plants on the twelfth–fourteenths days.
Further, the mentioned set of data was tested using the method of support vectors (Support Vector Machine).
Such kernel functions (kernels) as linear, polynomial, rbf and sigmoid were tested. The best result was demonstrated by
polynomial kernel.
As with neural networks, during using the support vector method on different days after herbicide application,
this method began to recognize well the difference between the experimental groups already on the seventh day after
herbicide application. Support vector method showed better result than used neural network. Already on the seventh day
after herbicide application, the maximum accuracy of recognition by the support vector method was almost 1,0 (100%),
while the used neural network provided only a maximum accuracy of 0,8.
In addition, the algorithm XGBoost (Extreme Gradient Boosting) was used on the mentioned data set – the popu-
lar machine-learning algorithm, which implements the gradient boosting model and is an alternative to regression meth-
ods and neural networks. This method creates a set of successively clarifying each other decision trees. The N-model of
tree is trained on the “errors” of set of n–1 models, and answers of models are weighted and summed. The implementa-
tion of the XGBClassifier algorithm in python language from the scikit-learn library was used. The XGBClassifier algo-
rithm was used for the task to recognize the influence of herbicide on different days after herbicide application without
data normalization. A thousand tests were conducted with different dividing into test and training samples. 30 % of
measured data of chlorophyll fluorescence induction were assigned to training sample. The algorithm shown the best re-
sults on classification on eleventh day after herbicide application.
The method to determine the soil humidity using the chlorophyll fluorescence induction was developed to help
an agronomist to determine the need for watering plants. For several weeks the chlorophyll fluorescence induction of
plants were measured, part of which was watered constantly, and the other part was not watered. Machine learning algo-
rithms had to determine the humidity level (regression task) by the curves of chlorophyll fluorescence induction. As in
the previous experiment, a neural network, a support vector method and XGBoost algorithm were used.
The used neural network contained three layers of neurons with a sigmoid activation function in the hidden lay-
ers and a linear activation function of neurons in the output layer. One of the results of regression of the neural network
is shown on Fig. 7.
Fig. 7. Results of neural network approximation of the dependence of the chlorophyll fluorescence induction curve on
soil moisture
The best result, obtained by the mentioned neural network, was R=0,81. The usage of support vector method
shown, that the best result was demonstrated by polynomial kernel of the forth degree with using of minimax normali-
zation of data. It was tested the possibility to give the different number of discretely measured points of curve of chlo-
rophyll fluorescence induction from 10 to maximum 90, which are measured by sensors and taken nonlinear by a power
scale (degree 1/8). The best result was obtained when using 70 points. The usage of XGBoost algorithm shown, that the
best results were obtained with using normalization of data by z-score (2). The regression is conducted with enough
high result even when taking discretely 10 points of curve of chlorophyll fluorescence induction by power scale. The re-
sults of all three machine-learning algorithms (regression coefficient and correlation coefficient) are shown in the ta-
ble. 1.
– matrix element in i row, k column;
Прикладне програмне забезпечення
where ikх – matrix element in i row, k column; ikх~ – normalized fluorescence value in i row, k column; kх –
average value of k column;
kх – standard deviation of chlorophyll fluorescence induction values in k column;. kmF , –
maximum value in k column, kF ,0 – minimum fluorescence value in k column.
The application of these methods of normalization made it possible to improve the average accuracy of recogni-
tion by neural network of groups of plants, treated with herbicide, on the seventh day after herbicide application from
66,7 % to 80,9 %. Neural networks provide an early determination of influence of stress factor on plant state even be-
fore appearance of visual signs on the plant. They showed good results on the seventh day after herbicide application,
while the visual signs of the herbicide influence only appear on the plants on the twelfth–fourteenths days.
Further, the mentioned set of data was tested using the method of support vectors (Support Vector Machine).
Such kernel functions (kernels) as linear, polynomial, rbf and sigmoid were tested. The best result was demonstrated by
polynomial kernel.
As with neural networks, during using the support vector method on different days after herbicide application,
this method began to recognize well the difference between the experimental groups already on the seventh day after
herbicide application. Support vector method showed better result than used neural network. Already on the seventh day
after herbicide application, the maximum accuracy of recognition by the support vector method was almost 1,0 (100%),
while the used neural network provided only a maximum accuracy of 0,8.
In addition, the algorithm XGBoost (Extreme Gradient Boosting) was used on the mentioned data set – the popu-
lar machine-learning algorithm, which implements the gradient boosting model and is an alternative to regression meth-
ods and neural networks. This method creates a set of successively clarifying each other decision trees. The N-model of
tree is trained on the “errors” of set of n–1 models, and answers of models are weighted and summed. The implementa-
tion of the XGBClassifier algorithm in python language from the scikit-learn library was used. The XGBClassifier algo-
rithm was used for the task to recognize the influence of herbicide on different days after herbicide application without
data normalization. A thousand tests were conducted with different dividing into test and training samples. 30 % of
measured data of chlorophyll fluorescence induction were assigned to training sample. The algorithm shown the best re-
sults on classification on eleventh day after herbicide application.
The method to determine the soil humidity using the chlorophyll fluorescence induction was developed to help
an agronomist to determine the need for watering plants. For several weeks the chlorophyll fluorescence induction of
plants were measured, part of which was watered constantly, and the other part was not watered. Machine learning algo-
rithms had to determine the humidity level (regression task) by the curves of chlorophyll fluorescence induction. As in
the previous experiment, a neural network, a support vector method and XGBoost algorithm were used.
The used neural network contained three layers of neurons with a sigmoid activation function in the hidden lay-
ers and a linear activation function of neurons in the output layer. One of the results of regression of the neural network
is shown on Fig. 7.
Fig. 7. Results of neural network approximation of the dependence of the chlorophyll fluorescence induction curve on
soil moisture
The best result, obtained by the mentioned neural network, was R=0,81. The usage of support vector method
shown, that the best result was demonstrated by polynomial kernel of the forth degree with using of minimax normali-
zation of data. It was tested the possibility to give the different number of discretely measured points of curve of chlo-
rophyll fluorescence induction from 10 to maximum 90, which are measured by sensors and taken nonlinear by a power
scale (degree 1/8). The best result was obtained when using 70 points. The usage of XGBoost algorithm shown, that the
best results were obtained with using normalization of data by z-score (2). The regression is conducted with enough
high result even when taking discretely 10 points of curve of chlorophyll fluorescence induction by power scale. The re-
sults of all three machine-learning algorithms (regression coefficient and correlation coefficient) are shown in the ta-
ble. 1.
– normalized fluorescence value in i row, k column;
Прикладне програмне забезпечення
where ikх – matrix element in i row, k column; ikх~ – normalized fluorescence value in i row, k column; kх –
average value of k column;
kх – standard deviation of chlorophyll fluorescence induction values in k column;. kmF , –
maximum value in k column, kF ,0 – minimum fluorescence value in k column.
The application of these methods of normalization made it possible to improve the average accuracy of recogni-
tion by neural network of groups of plants, treated with herbicide, on the seventh day after herbicide application from
66,7 % to 80,9 %. Neural networks provide an early determination of influence of stress factor on plant state even be-
fore appearance of visual signs on the plant. They showed good results on the seventh day after herbicide application,
while the visual signs of the herbicide influence only appear on the plants on the twelfth–fourteenths days.
Further, the mentioned set of data was tested using the method of support vectors (Support Vector Machine).
Such kernel functions (kernels) as linear, polynomial, rbf and sigmoid were tested. The best result was demonstrated by
polynomial kernel.
As with neural networks, during using the support vector method on different days after herbicide application,
this method began to recognize well the difference between the experimental groups already on the seventh day after
herbicide application. Support vector method showed better result than used neural network. Already on the seventh day
after herbicide application, the maximum accuracy of recognition by the support vector method was almost 1,0 (100%),
while the used neural network provided only a maximum accuracy of 0,8.
In addition, the algorithm XGBoost (Extreme Gradient Boosting) was used on the mentioned data set – the popu-
lar machine-learning algorithm, which implements the gradient boosting model and is an alternative to regression meth-
ods and neural networks. This method creates a set of successively clarifying each other decision trees. The N-model of
tree is trained on the “errors” of set of n–1 models, and answers of models are weighted and summed. The implementa-
tion of the XGBClassifier algorithm in python language from the scikit-learn library was used. The XGBClassifier algo-
rithm was used for the task to recognize the influence of herbicide on different days after herbicide application without
data normalization. A thousand tests were conducted with different dividing into test and training samples. 30 % of
measured data of chlorophyll fluorescence induction were assigned to training sample. The algorithm shown the best re-
sults on classification on eleventh day after herbicide application.
The method to determine the soil humidity using the chlorophyll fluorescence induction was developed to help
an agronomist to determine the need for watering plants. For several weeks the chlorophyll fluorescence induction of
plants were measured, part of which was watered constantly, and the other part was not watered. Machine learning algo-
rithms had to determine the humidity level (regression task) by the curves of chlorophyll fluorescence induction. As in
the previous experiment, a neural network, a support vector method and XGBoost algorithm were used.
The used neural network contained three layers of neurons with a sigmoid activation function in the hidden lay-
ers and a linear activation function of neurons in the output layer. One of the results of regression of the neural network
is shown on Fig. 7.
Fig. 7. Results of neural network approximation of the dependence of the chlorophyll fluorescence induction curve on
soil moisture
The best result, obtained by the mentioned neural network, was R=0,81. The usage of support vector method
shown, that the best result was demonstrated by polynomial kernel of the forth degree with using of minimax normali-
zation of data. It was tested the possibility to give the different number of discretely measured points of curve of chlo-
rophyll fluorescence induction from 10 to maximum 90, which are measured by sensors and taken nonlinear by a power
scale (degree 1/8). The best result was obtained when using 70 points. The usage of XGBoost algorithm shown, that the
best results were obtained with using normalization of data by z-score (2). The regression is conducted with enough
high result even when taking discretely 10 points of curve of chlorophyll fluorescence induction by power scale. The re-
sults of all three machine-learning algorithms (regression coefficient and correlation coefficient) are shown in the ta-
ble. 1.
– average
value of k column;
Прикладне програмне забезпечення
where ikх – matrix element in i row, k column; ikх~ – normalized fluorescence value in i row, k column; kх –
average value of k column;
kх – standard deviation of chlorophyll fluorescence induction values in k column;. kmF , –
maximum value in k column, kF ,0 – minimum fluorescence value in k column.
The application of these methods of normalization made it possible to improve the average accuracy of recogni-
tion by neural network of groups of plants, treated with herbicide, on the seventh day after herbicide application from
66,7 % to 80,9 %. Neural networks provide an early determination of influence of stress factor on plant state even be-
fore appearance of visual signs on the plant. They showed good results on the seventh day after herbicide application,
while the visual signs of the herbicide influence only appear on the plants on the twelfth–fourteenths days.
Further, the mentioned set of data was tested using the method of support vectors (Support Vector Machine).
Such kernel functions (kernels) as linear, polynomial, rbf and sigmoid were tested. The best result was demonstrated by
polynomial kernel.
As with neural networks, during using the support vector method on different days after herbicide application,
this method began to recognize well the difference between the experimental groups already on the seventh day after
herbicide application. Support vector method showed better result than used neural network. Already on the seventh day
after herbicide application, the maximum accuracy of recognition by the support vector method was almost 1,0 (100%),
while the used neural network provided only a maximum accuracy of 0,8.
In addition, the algorithm XGBoost (Extreme Gradient Boosting) was used on the mentioned data set – the popu-
lar machine-learning algorithm, which implements the gradient boosting model and is an alternative to regression meth-
ods and neural networks. This method creates a set of successively clarifying each other decision trees. The N-model of
tree is trained on the “errors” of set of n–1 models, and answers of models are weighted and summed. The implementa-
tion of the XGBClassifier algorithm in python language from the scikit-learn library was used. The XGBClassifier algo-
rithm was used for the task to recognize the influence of herbicide on different days after herbicide application without
data normalization. A thousand tests were conducted with different dividing into test and training samples. 30 % of
measured data of chlorophyll fluorescence induction were assigned to training sample. The algorithm shown the best re-
sults on classification on eleventh day after herbicide application.
The method to determine the soil humidity using the chlorophyll fluorescence induction was developed to help
an agronomist to determine the need for watering plants. For several weeks the chlorophyll fluorescence induction of
plants were measured, part of which was watered constantly, and the other part was not watered. Machine learning algo-
rithms had to determine the humidity level (regression task) by the curves of chlorophyll fluorescence induction. As in
the previous experiment, a neural network, a support vector method and XGBoost algorithm were used.
The used neural network contained three layers of neurons with a sigmoid activation function in the hidden lay-
ers and a linear activation function of neurons in the output layer. One of the results of regression of the neural network
is shown on Fig. 7.
Fig. 7. Results of neural network approximation of the dependence of the chlorophyll fluorescence induction curve on
soil moisture
The best result, obtained by the mentioned neural network, was R=0,81. The usage of support vector method
shown, that the best result was demonstrated by polynomial kernel of the forth degree with using of minimax normali-
zation of data. It was tested the possibility to give the different number of discretely measured points of curve of chlo-
rophyll fluorescence induction from 10 to maximum 90, which are measured by sensors and taken nonlinear by a power
scale (degree 1/8). The best result was obtained when using 70 points. The usage of XGBoost algorithm shown, that the
best results were obtained with using normalization of data by z-score (2). The regression is conducted with enough
high result even when taking discretely 10 points of curve of chlorophyll fluorescence induction by power scale. The re-
sults of all three machine-learning algorithms (regression coefficient and correlation coefficient) are shown in the ta-
ble. 1.
– standard deviation of chlorophyll fluorescence induction values in k column;
Прикладне програмне забезпечення
where ikх – matrix element in i row, k column; ikх~ – normalized fluorescence value in i row, k column; kх –
average value of k column;
kх – standard deviation of chlorophyll fluorescence induction values in k column;. kmF , –
maximum value in k column, kF ,0 – minimum fluorescence value in k column.
The application of these methods of normalization made it possible to improve the average accuracy of recogni-
tion by neural network of groups of plants, treated with herbicide, on the seventh day after herbicide application from
66,7 % to 80,9 %. Neural networks provide an early determination of influence of stress factor on plant state even be-
fore appearance of visual signs on the plant. They showed good results on the seventh day after herbicide application,
while the visual signs of the herbicide influence only appear on the plants on the twelfth–fourteenths days.
Further, the mentioned set of data was tested using the method of support vectors (Support Vector Machine).
Such kernel functions (kernels) as linear, polynomial, rbf and sigmoid were tested. The best result was demonstrated by
polynomial kernel.
As with neural networks, during using the support vector method on different days after herbicide application,
this method began to recognize well the difference between the experimental groups already on the seventh day after
herbicide application. Support vector method showed better result than used neural network. Already on the seventh day
after herbicide application, the maximum accuracy of recognition by the support vector method was almost 1,0 (100%),
while the used neural network provided only a maximum accuracy of 0,8.
In addition, the algorithm XGBoost (Extreme Gradient Boosting) was used on the mentioned data set – the popu-
lar machine-learning algorithm, which implements the gradient boosting model and is an alternative to regression meth-
ods and neural networks. This method creates a set of successively clarifying each other decision trees. The N-model of
tree is trained on the “errors” of set of n–1 models, and answers of models are weighted and summed. The implementa-
tion of the XGBClassifier algorithm in python language from the scikit-learn library was used. The XGBClassifier algo-
rithm was used for the task to recognize the influence of herbicide on different days after herbicide application without
data normalization. A thousand tests were conducted with different dividing into test and training samples. 30 % of
measured data of chlorophyll fluorescence induction were assigned to training sample. The algorithm shown the best re-
sults on classification on eleventh day after herbicide application.
The method to determine the soil humidity using the chlorophyll fluorescence induction was developed to help
an agronomist to determine the need for watering plants. For several weeks the chlorophyll fluorescence induction of
plants were measured, part of which was watered constantly, and the other part was not watered. Machine learning algo-
rithms had to determine the humidity level (regression task) by the curves of chlorophyll fluorescence induction. As in
the previous experiment, a neural network, a support vector method and XGBoost algorithm were used.
The used neural network contained three layers of neurons with a sigmoid activation function in the hidden lay-
ers and a linear activation function of neurons in the output layer. One of the results of regression of the neural network
is shown on Fig. 7.
Fig. 7. Results of neural network approximation of the dependence of the chlorophyll fluorescence induction curve on
soil moisture
The best result, obtained by the mentioned neural network, was R=0,81. The usage of support vector method
shown, that the best result was demonstrated by polynomial kernel of the forth degree with using of minimax normali-
zation of data. It was tested the possibility to give the different number of discretely measured points of curve of chlo-
rophyll fluorescence induction from 10 to maximum 90, which are measured by sensors and taken nonlinear by a power
scale (degree 1/8). The best result was obtained when using 70 points. The usage of XGBoost algorithm shown, that the
best results were obtained with using normalization of data by z-score (2). The regression is conducted with enough
high result even when taking discretely 10 points of curve of chlorophyll fluorescence induction by power scale. The re-
sults of all three machine-learning algorithms (regression coefficient and correlation coefficient) are shown in the ta-
ble. 1.
– maximum
value in k column,
Прикладне програмне забезпечення
where ikх – matrix element in i row, k column; ikх~ – normalized fluorescence value in i row, k column; kх –
average value of k column;
kх – standard deviation of chlorophyll fluorescence induction values in k column;. kmF , –
maximum value in k column, kF ,0 – minimum fluorescence value in k column.
The application of these methods of normalization made it possible to improve the average accuracy of recogni-
tion by neural network of groups of plants, treated with herbicide, on the seventh day after herbicide application from
66,7 % to 80,9 %. Neural networks provide an early determination of influence of stress factor on plant state even be-
fore appearance of visual signs on the plant. They showed good results on the seventh day after herbicide application,
while the visual signs of the herbicide influence only appear on the plants on the twelfth–fourteenths days.
Further, the mentioned set of data was tested using the method of support vectors (Support Vector Machine).
Such kernel functions (kernels) as linear, polynomial, rbf and sigmoid were tested. The best result was demonstrated by
polynomial kernel.
As with neural networks, during using the support vector method on different days after herbicide application,
this method began to recognize well the difference between the experimental groups already on the seventh day after
herbicide application. Support vector method showed better result than used neural network. Already on the seventh day
after herbicide application, the maximum accuracy of recognition by the support vector method was almost 1,0 (100%),
while the used neural network provided only a maximum accuracy of 0,8.
In addition, the algorithm XGBoost (Extreme Gradient Boosting) was used on the mentioned data set – the popu-
lar machine-learning algorithm, which implements the gradient boosting model and is an alternative to regression meth-
ods and neural networks. This method creates a set of successively clarifying each other decision trees. The N-model of
tree is trained on the “errors” of set of n–1 models, and answers of models are weighted and summed. The implementa-
tion of the XGBClassifier algorithm in python language from the scikit-learn library was used. The XGBClassifier algo-
rithm was used for the task to recognize the influence of herbicide on different days after herbicide application without
data normalization. A thousand tests were conducted with different dividing into test and training samples. 30 % of
measured data of chlorophyll fluorescence induction were assigned to training sample. The algorithm shown the best re-
sults on classification on eleventh day after herbicide application.
The method to determine the soil humidity using the chlorophyll fluorescence induction was developed to help
an agronomist to determine the need for watering plants. For several weeks the chlorophyll fluorescence induction of
plants were measured, part of which was watered constantly, and the other part was not watered. Machine learning algo-
rithms had to determine the humidity level (regression task) by the curves of chlorophyll fluorescence induction. As in
the previous experiment, a neural network, a support vector method and XGBoost algorithm were used.
The used neural network contained three layers of neurons with a sigmoid activation function in the hidden lay-
ers and a linear activation function of neurons in the output layer. One of the results of regression of the neural network
is shown on Fig. 7.
Fig. 7. Results of neural network approximation of the dependence of the chlorophyll fluorescence induction curve on
soil moisture
The best result, obtained by the mentioned neural network, was R=0,81. The usage of support vector method
shown, that the best result was demonstrated by polynomial kernel of the forth degree with using of minimax normali-
zation of data. It was tested the possibility to give the different number of discretely measured points of curve of chlo-
rophyll fluorescence induction from 10 to maximum 90, which are measured by sensors and taken nonlinear by a power
scale (degree 1/8). The best result was obtained when using 70 points. The usage of XGBoost algorithm shown, that the
best results were obtained with using normalization of data by z-score (2). The regression is conducted with enough
high result even when taking discretely 10 points of curve of chlorophyll fluorescence induction by power scale. The re-
sults of all three machine-learning algorithms (regression coefficient and correlation coefficient) are shown in the ta-
ble. 1.
– minimum fluorescence value in k column.
432
Прикладне програмне забезпечення
The application of these methods of normalization made it possible to improve the average accuracy of rec-
ognition by neural network of groups of plants, treated with herbicide, on the seventh day after herbicide application
from 66,7 % to 80,9 %. Neural networks provide an early determination of influence of stress factor on plant state even
before appearance of visual signs on the plant. They showed good results on the seventh day after herbicide applica-
tion, while the visual signs of the herbicide influence only appear on the plants on the twelfth–fourteenths days.
Further, the mentioned set of data was tested using the method of support vectors (Support Vector Machine). Such kernel
functions (kernels) as linear, polynomial, rbf and sigmoid were tested. The best result was demonstrated by polynomial kernel.
As with neural networks, during using the support vector method on different days after herbicide application,
this method began to recognize well the difference between the experimental groups already on the seventh day after
herbicide application. Support vector method showed better result than used neural network. Already on the seventh
day after herbicide application, the maximum accuracy of recognition by the support vector method was almost 1,0
(100%), while the used neural network provided only a maximum accuracy of 0,8.
In addition, the algorithm XGBoost (Extreme Gradient Boosting) was used on the mentioned data set – the
popular machine-learning algorithm, which implements the gradient boosting model and is an alternative to regres-
sion methods and neural networks. This method creates a set of successively clarifying each other decision trees. The
N-model of tree is trained on the “errors” of set of n–1 models, and answers of models are weighted and summed.
The implementation of the XGBClassifier algorithm in python language from the scikit-learn library was used. The
XGBClassifier algorithm was used for the task to recognize the influence of herbicide on different days after herbicide
application without data normalization. A thousand tests were conducted with different dividing into test and training
samples. 30 % of measured data of chlorophyll fluorescence induction were assigned to training sample. The algorithm
shown the best results on classification on eleventh day after herbicide application.
The method to determine the soil humidity using the chlorophyll fluorescence induction was developed to help
an agronomist to determine the need for watering plants. For several weeks the chlorophyll fluorescence induction of
plants were measured, part of which was watered constantly, and the other part was not watered. Machine learning
algorithms had to determine the humidity level (regression task) by the curves of chlorophyll fluorescence induction.
As in the previous experiment, a neural network, a support vector method and XGBoost algorithm were used.
The used neural network contained three layers of neurons with a sigmoid activation function in the hidden
layers and a linear activation function of neurons in the output layer. One of the results of regression of the neural
network is shown on Fig. 7.
Прикладне програмне забезпечення
where ikх – matrix element in i row, k column; ikх~ – normalized fluorescence value in i row, k column; kх –
average value of k column;
kх – standard deviation of chlorophyll fluorescence induction values in k column;. kmF , –
maximum value in k column, kF ,0 – minimum fluorescence value in k column.
The application of these methods of normalization made it possible to improve the average accuracy of recogni-
tion by neural network of groups of plants, treated with herbicide, on the seventh day after herbicide application from
66,7 % to 80,9 %. Neural networks provide an early determination of influence of stress factor on plant state even be-
fore appearance of visual signs on the plant. They showed good results on the seventh day after herbicide application,
while the visual signs of the herbicide influence only appear on the plants on the twelfth–fourteenths days.
Further, the mentioned set of data was tested using the method of support vectors (Support Vector Machine).
Such kernel functions (kernels) as linear, polynomial, rbf and sigmoid were tested. The best result was demonstrated by
polynomial kernel.
As with neural networks, during using the support vector method on different days after herbicide application,
this method began to recognize well the difference between the experimental groups already on the seventh day after
herbicide application. Support vector method showed better result than used neural network. Already on the seventh day
after herbicide application, the maximum accuracy of recognition by the support vector method was almost 1,0 (100%),
while the used neural network provided only a maximum accuracy of 0,8.
In addition, the algorithm XGBoost (Extreme Gradient Boosting) was used on the mentioned data set – the popu-
lar machine-learning algorithm, which implements the gradient boosting model and is an alternative to regression meth-
ods and neural networks. This method creates a set of successively clarifying each other decision trees. The N-model of
tree is trained on the “errors” of set of n–1 models, and answers of models are weighted and summed. The implementa-
tion of the XGBClassifier algorithm in python language from the scikit-learn library was used. The XGBClassifier algo-
rithm was used for the task to recognize the influence of herbicide on different days after herbicide application without
data normalization. A thousand tests were conducted with different dividing into test and training samples. 30 % of
measured data of chlorophyll fluorescence induction were assigned to training sample. The algorithm shown the best re-
sults on classification on eleventh day after herbicide application.
The method to determine the soil humidity using the chlorophyll fluorescence induction was developed to help
an agronomist to determine the need for watering plants. For several weeks the chlorophyll fluorescence induction of
plants were measured, part of which was watered constantly, and the other part was not watered. Machine learning algo-
rithms had to determine the humidity level (regression task) by the curves of chlorophyll fluorescence induction. As in
the previous experiment, a neural network, a support vector method and XGBoost algorithm were used.
The used neural network contained three layers of neurons with a sigmoid activation function in the hidden lay-
ers and a linear activation function of neurons in the output layer. One of the results of regression of the neural network
is shown on Fig. 7.
Fig. 7. Results of neural network approximation of the dependence of the chlorophyll fluorescence induction curve on
soil moisture
The best result, obtained by the mentioned neural network, was R=0,81. The usage of support vector method
shown, that the best result was demonstrated by polynomial kernel of the forth degree with using of minimax normali-
zation of data. It was tested the possibility to give the different number of discretely measured points of curve of chlo-
rophyll fluorescence induction from 10 to maximum 90, which are measured by sensors and taken nonlinear by a power
scale (degree 1/8). The best result was obtained when using 70 points. The usage of XGBoost algorithm shown, that the
best results were obtained with using normalization of data by z-score (2). The regression is conducted with enough
high result even when taking discretely 10 points of curve of chlorophyll fluorescence induction by power scale. The re-
sults of all three machine-learning algorithms (regression coefficient and correlation coefficient) are shown in the ta-
ble. 1.
Fig. 7. Results of neural network approximation of the dependence of the chlorophyll fluorescence induction
curve on soil moisture
The best result, obtained by the mentioned neural network, was R=0,81. The usage of support vector method
shown, that the best result was demonstrated by polynomial kernel of the forth degree with using of minimax nor-
malization of data. It was tested the possibility to give the different number of discretely measured points of curve
of chlorophyll fluorescence induction from 10 to maximum 90, which are measured by sensors and taken nonlinear
by a power scale (degree 1/8). The best result was obtained when using 70 points. The usage of XGBoost algorithm
shown, that the best results were obtained with using normalization of data by z-score (2). The regression is con-
ducted with enough high result even when taking discretely 10 points of curve of chlorophyll fluorescence induction
by power scale. The results of all three machine-learning algorithms (regression coefficient and correlation coef-
ficient) are shown in the table. 1.
Table1. The best results, obtained by different regression algorithms
Regression algorithm
Прикладне програмне забезпечення
Table1. The best results, obtained by different regression algorithms
Regression algorithm 2
maxR maxR
Neural network 0,65 0,81
Support vector method 0,54 0,73
XGBoost algorithms 0,52 0,72
Testing the correctness of the operation of applied software of network nodes
Acquisition of reliable measured data and making the proper managerial decisions depend on reliability and sta-
bility of operation of measuring sensors, data transfer channels and proper applied software, which controls their work.
Since, in our applied task the measurements are made by autonomous measuring nodes, which organize into a wireless
sensor network, so the correctness of operation of applied software greatly influences energy efficiency of autonomous
nodes in whole. If the applied software and data transfer protocols are developed and configured incorrectly, then net-
work nodes will quickly discharge, which make it impossible for such wireless network to fulfil its target task – meas-
urement and acquisition of measured data from certain territory of agricultural lands during a certain time with a pur-
pose to make proper managerial decision to help the user.
That is why the testing of the operation of such network in combination of its hardware and software in field
conditions plays an important role for further scaling of such network and its implementation in real applications. Since
the deployment of the whole network for testing in field conditions is a rather long and expensive process, as a rule, on-
ly a cluster of such network is tested, what makes it possible to discover weaknesses and miscalculations in develop-
ment both applied software and hardware.
The proposed network is intended for needs of digital agriculture for usage on the large territories of agricultural
lands. The purpose of the network defines its characteristics, which are very important during testing. They include pro-
tection from climatic conditions, duration of autonomous operation without replacing the power elements and conduct-
ing maintenance, high reliability of operation, low power consumption, sufficient range of wireless data transmission.
The main parameters, which determine the correctness of applied software operation and have to be evaluated
during testing, are:
1) The time of self-organization – time, which is needed to connect to network all nodes.
2) The range of communication – the distance, at which there is a stable connection between nodes.
3) The quality of communication – refers to the ratio of successfully received messages by a node to the total
number of messages sent to this node.
4) Time of autonomous operation without replacing the power elements.
Firstly, testing of the communication range and reliability of message transmission, or, in other words, the quali-
ty of the wireless channel between the wireless sensor and the network coordinator, was performed. Tests were con-
ducted for different environmental conditions, in particular for four types of terrain:
– open territory, where the line of sight between signal source and signal receiver was present;
– low vegetation, grass, many bushes, the line of sight between signal source and signal receiver was present;
– dense vegetation, many high trees, line of sight between signal source and signal receiver was absent in most
cases;
– urban territory, line of sight between signal source and receiver was almost always absent.
To conduct testing a special applied software for wireless sensor and network coordinator was created. Testing
was conducted accordingly to the next order:
1. The wireless measuring node and coordinator sequentially were removed each from other on the distance, di-
visible by 10 meters, until 160 meters.
2. The wireless measuring node transmitted total 100 messages, 1 message every 500 milliseconds. The size of
every message equalled 1 byte. Coordinator recorded the number of received messages and did not send the acknowl-
edgment of message receiving.
3. The coordinator transmitted total 100 messages, 1 message every 500 milliseconds. The size of every message
equalled 1 byte. The wireless measuring node recorded the number of received messages and did not send the acknowl-
edgment of message receiving.
4. Scenario items 2 and 3 were repeated three times for every distance. This number of repetitions was necessary
to avoid the influence of accidental temporary obstacles or noise, which could be caused by different sources.
5. Coefficient of communication reliability (quality) was calculated as number of received messages, divided by
number of transmitted messages, and then this result was ranged from 0 to 100 percent. The average of six measure-
ments of communication quality for each of given distances was used for plotting the graph.
By results of testing, the general dependence of coefficient of communication reliability (quality) on distance be-
tween network coordinator and wireless measuring node was built (Fig. 8).
The graph (Fig. 8) shows, that distances of stable connection for an open territory and territory with low vegeta-
tion almost coincide. During testing the wireless network in conditions of dense vegetation or urban or rural buildings
the reliability (quality) of connection significantly worsens because of the absence of direct sight between the coordina-
tor and wireless measuring node.
The territory of orchard (cherry trees) was chosen as place for conducting the natural experiment, which is the
closest to the conditions of target application. The network cluster consisted of network coordinator and four wireless
Прикладне програмне забезпечення
Table1. The best results, obtained by different regression algorithms
Regression algorithm 2
maxR maxR
Neural network 0,65 0,81
Support vector method 0,54 0,73
XGBoost algorithms 0,52 0,72
Testing the correctness of the operation of applied software of network nodes
Acquisition of reliable measured data and making the proper managerial decisions depend on reliability and sta-
bility of operation of measuring sensors, data transfer channels and proper applied software, which controls their work.
Since, in our applied task the measurements are made by autonomous measuring nodes, which organize into a wireless
sensor network, so the correctness of operation of applied software greatly influences energy efficiency of autonomous
nodes in whole. If the applied software and data transfer protocols are developed and configured incorrectly, then net-
work nodes will quickly discharge, which make it impossible for such wireless network to fulfil its target task – meas-
urement and acquisition of measured data from certain territory of agricultural lands during a certain time with a pur-
pose to make proper managerial decision to help the user.
That is why the testing of the operation of such network in combination of its hardware and software in field
conditions plays an important role for further scaling of such network and its implementation in real applications. Since
the deployment of the whole network for testing in field conditions is a rather long and expensive process, as a rule, on-
ly a cluster of such network is tested, what makes it possible to discover weaknesses and miscalculations in develop-
ment both applied software and hardware.
The proposed network is intended for needs of digital agriculture for usage on the large territories of agricultural
lands. The purpose of the network defines its characteristics, which are very important during testing. They include pro-
tection from climatic conditions, duration of autonomous operation without replacing the power elements and conduct-
ing maintenance, high reliability of operation, low power consumption, sufficient range of wireless data transmission.
The main parameters, which determine the correctness of applied software operation and have to be evaluated
during testing, are:
1) The time of self-organization – time, which is needed to connect to network all nodes.
2) The range of communication – the distance, at which there is a stable connection between nodes.
3) The quality of communication – refers to the ratio of successfully received messages by a node to the total
number of messages sent to this node.
4) Time of autonomous operation without replacing the power elements.
Firstly, testing of the communication range and reliability of message transmission, or, in other words, the quali-
ty of the wireless channel between the wireless sensor and the network coordinator, was performed. Tests were con-
ducted for different environmental conditions, in particular for four types of terrain:
– open territory, where the line of sight between signal source and signal receiver was present;
– low vegetation, grass, many bushes, the line of sight between signal source and signal receiver was present;
– dense vegetation, many high trees, line of sight between signal source and signal receiver was absent in most
cases;
– urban territory, line of sight between signal source and receiver was almost always absent.
To conduct testing a special applied software for wireless sensor and network coordinator was created. Testing
was conducted accordingly to the next order:
1. The wireless measuring node and coordinator sequentially were removed each from other on the distance, di-
visible by 10 meters, until 160 meters.
2. The wireless measuring node transmitted total 100 messages, 1 message every 500 milliseconds. The size of
every message equalled 1 byte. Coordinator recorded the number of received messages and did not send the acknowl-
edgment of message receiving.
3. The coordinator transmitted total 100 messages, 1 message every 500 milliseconds. The size of every message
equalled 1 byte. The wireless measuring node recorded the number of received messages and did not send the acknowl-
edgment of message receiving.
4. Scenario items 2 and 3 were repeated three times for every distance. This number of repetitions was necessary
to avoid the influence of accidental temporary obstacles or noise, which could be caused by different sources.
5. Coefficient of communication reliability (quality) was calculated as number of received messages, divided by
number of transmitted messages, and then this result was ranged from 0 to 100 percent. The average of six measure-
ments of communication quality for each of given distances was used for plotting the graph.
By results of testing, the general dependence of coefficient of communication reliability (quality) on distance be-
tween network coordinator and wireless measuring node was built (Fig. 8).
The graph (Fig. 8) shows, that distances of stable connection for an open territory and territory with low vegeta-
tion almost coincide. During testing the wireless network in conditions of dense vegetation or urban or rural buildings
the reliability (quality) of connection significantly worsens because of the absence of direct sight between the coordina-
tor and wireless measuring node.
The territory of orchard (cherry trees) was chosen as place for conducting the natural experiment, which is the
closest to the conditions of target application. The network cluster consisted of network coordinator and four wireless
Neural network 0,65 0,81
Support vector method 0,54 0,73
XGBoost algorithms 0,52 0,72
433
Прикладне програмне забезпечення
Testing the correctness of the operation of applied software of network nodes
Acquisition of reliable measured data and making the proper managerial decisions depend on reliability and
stability of operation of measuring sensors, data transfer channels and proper applied software, which controls their
work. Since, in our applied task the measurements are made by autonomous measuring nodes, which organize into a
wireless sensor network, so the correctness of operation of applied software greatly influences energy efficiency of au-
tonomous nodes in whole. If the applied software and data transfer protocols are developed and configured incorrectly,
then network nodes will quickly discharge, which make it impossible for such wireless network to fulfil its target task
– measurement and acquisition of measured data from certain territory of agricultural lands during a certain time with
a purpose to make proper managerial decision to help the user.
That is why the testing of the operation of such network in combination of its hardware and software in field
conditions plays an important role for further scaling of such network and its implementation in real applications. Since
the deployment of the whole network for testing in field conditions is a rather long and expensive process, as a rule,
only a cluster of such network is tested, what makes it possible to discover weaknesses and miscalculations in develop-
ment both applied software and hardware.
The proposed network is intended for needs of digital agriculture for usage on the large territories of agricultural
lands. The purpose of the network defines its characteristics, which are very important during testing. They include pro-
tection from climatic conditions, duration of autonomous operation without replacing the power elements and conducting
maintenance, high reliability of operation, low power consumption, sufficient range of wireless data transmission.
The main parameters, which determine the correctness of applied software operation and have to be evaluated
during testing, are:
1) The time of self-organization – time, which is needed to connect to network all nodes.
2) The range of communication – the distance, at which there is a stable connection between nodes.
3) The quality of communication – refers to the ratio of successfully received messages by a node to the total
number of messages sent to this node.
4) Time of autonomous operation without replacing the power elements.
Firstly, testing of the communication range and reliability of message transmission, or, in other words, the
quality of the wireless channel between the wireless sensor and the network coordinator, was performed. Tests were
conducted for different environmental conditions, in particular for four types of terrain:
– open territory, where the line of sight between signal source and signal receiver was present;
– low vegetation, grass, many bushes, the line of sight between signal source and signal receiver was present;
– dense vegetation, many high trees, line of sight between signal source and signal receiver was absent in
most cases;
– urban territory, line of sight between signal source and receiver was almost always absent.
To conduct testing a special applied software for wireless sensor and network coordinator was created. Testing
was conducted accordingly to the next order:
1. The wireless measuring node and coordinator sequentially were removed each from other on the distance,
divisible by 10 meters, until 160 meters.
2. The wireless measuring node transmitted total 100 messages, 1 message every 500 milliseconds. The size of
every message equalled 1 byte. Coordinator recorded the number of received messages and did not send the acknowl-
edgment of message receiving.
3. The coordinator transmitted total 100 messages, 1 message every 500 milliseconds. The size of every mes-
sage equalled 1 byte. The wireless measuring node recorded the number of received messages and did not send the
acknowledgment of message receiving.
4. Scenario items 2 and 3 were repeated three times for every distance. This number of repetitions was neces-
sary to avoid the influence of accidental temporary obstacles or noise, which could be caused by different sources.
5. Coefficient of communication reliability (quality) was calculated as number of received messages, divided
by number of transmitted messages, and then this result was ranged from 0 to 100 percent. The average of six measure-
ments of communication quality for each of given distances was used for plotting the graph.
By results of testing, the general dependence of coefficient of communication reliability (quality) on distance
between network coordinator and wireless measuring node was built (Fig. 8).
The graph (Fig. 8) shows, that distances of stable connection for an open territory and territory with low
vegetation almost coincide. During testing the wireless network in conditions of dense vegetation or urban or rural
buildings the reliability (quality) of connection significantly worsens because of the absence of direct sight between
the coordinator and wireless measuring node.
The territory of orchard (cherry trees) was chosen as place for conducting the natural experiment, which is the
closest to the conditions of target application. The network cluster consisted of network coordinator and four wireless
measuring nodes. The operation of network coordinator and measuring nodes was in normal mode accordingly to users
guide. Measuring nodes was placed on the height of 1,5 meters above ground surface and at distance of 20–60 meters
from the coordinator. The average time of network organization was 6,75 seconds. A total of 82 measurement sessions
were conducted. The total number of successfully transferred measurements was 76. Accordingly, the integral score of
unsuccessfully transferred measurements equalled 7,32 %, what was caused by random noise and obstacles on the sig-
nal path. The average level of battery discharge for all measurement sessions was 25–40%, depending on the distance
between measuring nodes and coordinator and presence of noise and obstacles.
434
Прикладне програмне забезпечення
Прикладне програмне забезпечення
measuring nodes. The operation of network coordinator and measuring nodes was in normal mode accordingly to users
guide. Measuring nodes was placed on the height of 1,5 meters above ground surface and at distance of 20–60 meters
from the coordinator. The average time of network organization was 6,75 seconds. A total of 82 measurement sessions
were conducted. The total number of successfully transferred measurements was 76. Accordingly, the integral score of
unsuccessfully transferred measurements equalled 7,32 %, what was caused by random noise and obstacles on the sig-
nal path. The average level of battery discharge for all measurement sessions was 25–40%, depending on the distance
between measuring nodes and coordinator and presence of noise and obstacles.
Fig. 8. The general dependence of coefficient of communication reliability (quality) on distance between network coor-
dinator and wireless measuring node
Obtained data of testing prove the reasonability of using the wireless sensor networks in the first chain of data
acquisition and making-decision support systems, which are intended for usage in agriculture, ecological monitoring
and environmental protection. The correctness of operation of applied software of network nodes and their interaction
within whole network approved by fast self-organization of network, a relatively high number of successfully conduct-
ed and transferred measurements, as well as permissible discharge of batteries. A small modernization of hardware and
applied software of network nodes, which will be made by results of testing, make it possible to get in the future the
higher exploitation features of whole system for target applications, not only for agriculture and ecological monitoring,
but also for another fields of human activity, e.g. for controlling the quality of food products [12].
Conclusion
The article describes applied software of units of such complex hardware-software system, as plants` state moni-
toring system for application in agriculture and ecological monitoring. The mentioned system consists of data acquisi-
tion system in the form of wireless sensor network and adaptive part in the form of decision-making support system.
The authors described main applied software of autonomous nodes of wireless sensor network and implementation of
some program functions of decision-making support system. Wireless sensor network includes many autonomous wire-
less sensors, so the main criteria during applied software creation was assuring the energy efficiency of operation of au-
tonomous measuring nodes and network coordinator, and correct interaction of nodes within all network. The main pa-
rameters, which define the correctness of applied software operation, were estimated during conducted testing. These
parameters include, for example, time of network self-organization, distance and quality of stable communication, time
of autonomous operation of wireless nodes without charging and so on. Obtained data of testing of network nodes in
field conditions prove the reasonability of using the wireless sensor networks as data acquisition system for making-
decision support system to help the agronomist.
To develop some software function of decision-making support system, firstly, the authors have developed the
applied method for express-diagnostics of plant state and estimating factors, which influence the plant. For this purpose,
the natural experiments were conducted to determine the sufficient dose of herbicide application and estimating the soil
moisture using the chlorophyll fluorescence induction. For processing measured data, several methods of machine
learning were used, including neural network approach. Application of machine learning methods made it possible, on
the base of acquired data, to make early diagnostics of influence of stress factors on the plant even before the appear-
ance of visual manifestations of such negative influence and determine the decrease of soil moisture through the diag-
nostics of plant itself, and inform the user about this. Thus, it can be claimed, that the proposed system possesses some
functions of artificial intelligence thanks to the developed applied software.
References
1. Dolphin Engineering Sagl (2022) PreDiVine. [Online] Available from: https://www.dolphin-engineering.ch [Accessed: 31 July 2022].
2. AgroSense (2022). [Online] Available from: http://www.agrosense.hu [Accessed: 31 July 2022].
3. Libelium (2012) Wireless Sensor Networks to monitor food sustainability. [Online] Available from:
https://www.libelium.com/libeliumworld/food_sustainability_monitoring_sensor_network/ [Accessed: 31 July 2022].
4. Mendez, G.R., Yunus, M.A.M., Mukhopadhyay, S.C. (2012) A WiFi based smart wireless sensor network for monitoring an agricultural envi-
ronment. in Instrumentation and Measurement Technology Conference (I2MTC). Graz, Austria, 2012. pp. 2640-2645.
Fig. 8. The general dependence of coefficient of communication reliability (quality) on distance between network co-
ordinator and wireless measuring node
Obtained data of testing prove the reasonability of using the wireless sensor networks in the first chain
of data acquisition and making-decision support systems, which are intended for usage in agriculture, ecological
monitoring and environmental protection. The correctness of operation of applied software of network nodes and
their interaction within whole network approved by fast self-organization of network, a relatively high number
of successfully conducted and transferred measurements, as well as permissible discharge of batteries. A small
modernization of hardware and applied software of network nodes, which will be made by results of testing, make
it possible to get in the future the higher exploitation features of whole system for target applications, not only
for agriculture and ecological monitoring, but also for another fields of human activity, e.g. for controlling the
quality of food products [12].
Conclusion
The article describes applied software of units of such complex hardware-software system, as plants`
state monitoring system for application in agriculture and ecological monitoring. The mentioned system consists
of data acquisition system in the form of wireless sensor network and adaptive part in the form of decision-
making support system. The authors described main applied software of autonomous nodes of wireless sensor
network and implementation of some program functions of decision-making support system. Wireless sensor
network includes many autonomous wireless sensors, so the main criteria during applied software creation was
assuring the energy efficiency of operation of autonomous measuring nodes and network coordinator, and correct
interaction of nodes within all network. The main parameters, which define the correctness of applied software
operation, were estimated during conducted testing. These parameters include, for example, time of network
self-organization, distance and quality of stable communication, time of autonomous operation of wireless nodes
without charging and so on. Obtained data of testing of network nodes in field conditions prove the reasonability
of using the wireless sensor networks as data acquisition system for making-decision support system to help the
agronomist.
To develop some software function of decision-making support system, firstly, the authors have devel-
oped the applied method for express-diagnostics of plant state and estimating factors, which influence the plant.
For this purpose, the natural experiments were conducted to determine the sufficient dose of herbicide applica-
tion and estimating the soil moisture using the chlorophyll fluorescence induction. For processing measured data,
several methods of machine learning were used, including neural network approach. Application of machine
learning methods made it possible, on the base of acquired data, to make early diagnostics of influence of stress
factors on the plant even before the appearance of visual manifestations of such negative influence and determine
the decrease of soil moisture through the diagnostics of plant itself, and inform the user about this. Thus, it can
be claimed, that the proposed system possesses some functions of artificial intelligence thanks to the developed
applied software.
References
1. Dolphin Engineering Sagl (2022) PreDiVine. [Online] Available from: https://www.dolphin-engineering.ch [Accessed: 31 July 2022].
2. AgroSense (2022). [Online] Available from: http://www.agrosense.hu [Accessed: 31 July 2022].
3. Libelium (2012) Wireless Sensor Networks to monitor food sustainability. [Online] Available from: https://www.libelium.com/libeliumworld/
food_sustainability_monitoring_sensor_network/ [Accessed: 31 July 2022].
435
Прикладне програмне забезпечення
4. Mendez, G.R., Yunus, M.A.M., Mukhopadhyay, S.C. (2012) A WiFi based smart wireless sensor network for monitoring an agricultural
environment. in Instrumentation and Measurement Technology Conference (I2MTC). Graz, Austria, 2012. pp. 2640-2645.
5. Romanov, V., Galelyuka, I., Antonova, H., Kovyrova, O., Hrusha, V., Voronenko, O. (2019) Application of wireless sensor networks for
digital agriculture. Intelligent Data Acquisition and Advanced Computing Systems: Technology and Applications, IDAACS’2019. Metz,
France. 2019, September 18–21. pp. 340-344.
6. Chlorophyll fluorescence spectral discrimination by artificial neural network methods (2002). DEFRA project code HH1530SPC.
2002.
7. Kirova, M., Ceppi, G., Chernev, P., Goltsev, V., Strasser, R. (2009) Using Artificial Neural Networks for Plant Taxonomic Determina-
tion Based on Chlorophyll Fluorescence Induction Curves. Biotechnology and Biotechnological Equipment. Volume 23, 2009. Issue
sup1: XI Anniversary scientific conference. pp. 941-946. https://doi.org/10.1080/13102818.2009.10818577
8. Goltsev, V., Zaharieva, I., Chernev, P., Kouzmanova, M., Kalaji, H.M., Yordanov, I., Krasteva, V., Alexandrov, V., Stefanov, D., Allkh-
verdiev, S.I., Strasser, R.J. (2012) Drought-induced modification of photosynthetic electron transport in intact leaves: Analysis and use
of neural network as a tool for a rapid non-invasive estimation. Biochimica et Biophysica Acta. 1817(8). 2012. pp. 1490–1498.
9. Rybka, K., Janaszek-Mankowska, M., Siedlarz, P., Mankowski, D. (2019) Machine learning in determination of water saturation defi-
cit in wheat leaves on basis of Chl a fluorescence parameters. Photosynthetica. 57 (1). 2019. pp. 226-230. https://doi.org/10.32615/
ps.2019.017
10. Soja, G., Soja, A.M. (2005) Recognizing the Sources of Stress in Wheat and Bean by using chlorophyll fluorescence induction param-
eters as inputs for neural network models. Phyton. Special issue: «D. Grill». 45 (3). 2005. pp. 157-168.
11. Pantazi, X.E., Moshou, D., Kasampalis, D. and Tsouvaltzis, P. (2014) Automatic Accessment of Phenotypes in lettuce plants by using
Chlorophyll Fluorescence Kinetics and Machine Learning. International Conference of Agricultural Engineering. 2014. Zurich. 6-10
July 2014. pp.167-176.
12. Romanov, V., Galelyuka, I., Voronenko, O. (2021) WSN for Food Product Quality Control. Intelligent Data Acquisition and Advanced
Computing Systems: Technology and Applications, IDAACS’2021. Cracow, Poland. 2021, September 22–25. pp. 580–583.
Received 03.09.2022
About authors:
Romanov Volodymyr Oleksandrovych,
Doctor in Computer Sciences,
Head of department.
Number of publications in domestic issues – 201.
Number of publications in foreign issues – 54.
Hirsch index (Scopus) – 5.
Hirsch index (Google Scholar) – 11.
http://orcid.org/ 0000-0001-6277-8756.
Antonova Hanna Valeriivna,
Junior researcher.
Number of publications in domestic issues – 25.
Number of publications in foreign issues – 7.
Hirsch index (Google Scholar) – 3.
http://orcid.org/0000-0002-1876-8267.
Galelyuka Igor Bogdanovych,
PhD in Computer Sciences,
Leading researcher.
Number of publications in domestic issues – 92.
Number of publications in foreign issues – 43.
Hirsch index (Scopus) – 5.
Hirsch index (Google Scholar) – 10.
http://orcid.org/0000-0003-1504-4439.
Hrusha Volodymyr Myhaylovych,
PhD in Computer Sciences,
Junior researcher.
Number of publications in domestic issues – 41.
Number of publications in foreign issues – 11.
Hirsch index (Scopus) – 3.
Hirsch index (Google Scholar) – 6.
http://orcid.org/0000-0003-1504-4439.
436
Прикладне програмне забезпечення
Kedych Anna Vasylivna,
Leading engineer-programmer.
Number of publications in domestic issues – 3.
Hirsch index (Google Scholar) – 1.
https://orcid.org/0000-0003-1784-4296.
Voronenko Oleksandr Volodymyrovych,
Junior researcher.
Number of publications in domestic issues – 19.
Number of publications in foreign issues – 6.
Hirsch index (Scopus) – 2.
Hirsch index (Google Scholar) – 3.
http://orcid.org/0000-0002-5022-8017.
Place of work:
Glushkov` Institute of Cybernetics of NAS of Ukraine,
03187, Kyiv, prospect Akademika Glushkova, 40.
Tel.: +38 (044) 526-32-04
E-mail: VRomanov@i.ua, galelyuka@gmail.com
Прізвища та ініціали авторів і назва доповіді українською мовою:
Романов В.О., Антонова Г.В., Галелюка І.Б., Груша В.М.,
Кедич А.В., Вороненко О.В.
Розробка та перевірка програмних засобів мережі
бездротових сенсорів для сільського господарства
Прізвища та ініціали авторів і назва доповіді англійською мовою:
Romanov V.O., Antonova H.V., Galelyuka I.B., Hrusha V.M.,
Kedych A.V., Voronenko O.V.
Creation and test of applied software of network
of wireless sensors for agriculture
Контакти для редактора: Галелюка Ігор Богданович,
провідний науковий співробітник Інституту кібернетики
імені В.М. Глушкова НАН України,
e-mail: galelyuka@gmail.com,
тел.: (38)(044) 526-32-04, (38)(050) 986-90-33
|