Новий модифікований алгоритм ядра fuzzy c-means, що використовується для виявлення плям на листу бавовника
Image segmentation is a significant and difficult subject that is a prerequisite for both basic image analysis and sophisticated picture interpretation. In image analysis, picture segmentation is crucial. Several different applications, including those related to medicine, facial identification, Cot...
Saved in:
| Date: | 2023 |
|---|---|
| Main Authors: | , |
| Format: | Article |
| Language: | English |
| Published: |
The National Technical University of Ukraine "Igor Sikorsky Kyiv Polytechnic Institute"
2023
|
| Subjects: | |
| Online Access: | https://journal.iasa.kpi.ua/article/view/297405 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Journal Title: | System research and information technologies |
| Download file: |  |
Institution
System research and information technologies| _version_ | 1867334441094873088 |
|---|---|
| author | Paithane, Pradip Wagh, Sarita Jibhau |
| author_facet | Paithane, Pradip Wagh, Sarita Jibhau |
| author_institution_txt_mv | [
{
"author": "Pradip Paithane",
"institution": "Vidya Pratishthan’s Kamalnayan Bajaj Institute of Engineering and Technology, Pune"
},
{
"author": "Sarita Jibhau Wagh",
"institution": "T.C. College Baramati, Pune"
}
] |
| author_sort | Paithane, Pradip |
| baseUrl_str | http://journal.iasa.kpi.ua/oai |
| collection | OJS |
| datestamp_date | 2024-02-01T21:03:07Z |
| description | Image segmentation is a significant and difficult subject that is a prerequisite for both basic image analysis and sophisticated picture interpretation. In image analysis, picture segmentation is crucial. Several different applications, including those related to medicine, facial identification, Cotton disease diagnosis, and map object detection, benefit from image segmentation. In order to segment images, the clustering approach is used. The two types of clustering algorithms are Crisp and Fuzzy. Crisp clustering is superior to fuzzy clustering. Fuzzy clustering uses the well-known FCM approach to enhance the results of picture segmentation. KFCM technique for image segmentation can be utilized to overcome FCM’s shortcomings in noisy and nonlinear separable images. In the KFCM approach, the Gaussian kernel function transforms high-dimensional, nonlinearly separable data into linearly separable data before applying FCM to the data. KFCM is enhancing noisy picture segmentation results. KFCM increases the accuracy rate but ignores neighboring pixels. The Modified Kernel Fuzzy C-Means approach is employed to get over this problem. The NMKFCM approach enhances picture segmentation results by including neighboring pixel information into the objective function. This suggested technique is used to find “blackarm” spots on cotton leaves. A fungal leaf disease called “blackarm” leaf spot results in brown leaves with purple borders. The bacterium can harm cotton plants, causing angular leaf blotches that range in color from red to brown. |
| doi_str_mv | 10.20535/SRIT.2308-8893.2023.4.07 |
| first_indexed | 2025-07-17T10:28:26Z |
| format | Article |
| fulltext |
Pradip M. Paithane, Sarita Jibhau Wagh, 2023
Системні дослідження та інформаційні технології, 2023, № 4 85
UDC 62-50
DOI: 10.20535/SRIT.2308-8893.2023.4.07
NOVEL MODIFIED KERNEL FUZZY C-MEANS ALGORITHM
USED FOR COTTON LEAF SPOT DETECTION
PRADIP M. PAITHANE, SARITA JIBHAU WAGH
Abstract. Image segmentation is a significant and difficult subject that is a prerequi-
site for both basic image analysis and sophisticated picture interpretation. In image
analysis, picture segmentation is crucial. Several different applications, including
those related to medicine, facial identification, Cotton disease diagnosis, and map
object detection, benefit from image segmentation. In order to segment images, the
clustering approach is used. The two types of clustering algorithms are Crisp and
Fuzzy. Crisp clustering is superior to fuzzy clustering. Fuzzy clustering uses the
well-known FCM approach to enhance the results of picture segmentation. KFCM
technique for image segmentation can be utilized to overcome FCM’s shortcomings
in noisy and nonlinear separable images. In the KFCM approach, the Gaussian ker-
nel function transforms high-dimensional, nonlinearly separable data into linearly
separable data before applying FCM to the data. KFCM is enhancing noisy picture
segmentation results. KFCM increases the accuracy rate but ignores neighboring
pixels. The Modified Kernel Fuzzy C-Means approach is employed to get over this
problem. The NMKFCM approach enhances picture segmentation results by includ-
ing neighboring pixel information into the objective function. This suggested tech-
nique is used to find “blackarm” spots on cotton leaves. A fungal leaf disease called
“blackarm” leaf spot results in brown leaves with purple borders. The bacterium can
harm cotton plants, causing angular leaf blotches that range in color from red to
brown.
Keywords: Cluster Accuracy Rate (CAR), Clustering, Cotton Leaf Disease, Fuzzy
Clustering Method (FCM), Kernel Fuzzy C-means Algorithm (KFCM), Novel
Modified Kernel Fuzzy C-Means Clustering Algorithm (NMKFCM).
INTRODUCTION
Cotton is the most significant cash crop farmed in Maharashtra, India. The pri-
mary issue reducing cotton output is disease on the plant. Because a minute dif-
ference in color pattern might be caused by a different disease that is present on a
cotton leaf, we know that the human eye’s perception is not powerful enough to
enable it to recognize minute variations in the diseased region of an image. The
cotton plant’s leaf is the disease’s primary source. The leaves of the cotton plant
are where 80–90% of the illness is located. One crucial technique for separating a
picture into its backdrop and its objects is image segmentation. Clustering is one
of the crucial phases in picture segmentation. In the early portion of the season,
affected crops may develop slowly or be stunted. Blackening of the roots is
a symptom of the illness, which results in the destruction of the root cortex (outer
layer). Thielaviopsis basicola does not kill seedlings on its own, however some
roots may perish. Significant black root rot exposes the root to Pythium or
Rhizoctonia infection. When growth begins in warmer temperatures, the dead
Pradip M. Paithane, Sarita Jibhau Wagh
ISSN 1681–6048 System Research & Information Technologies, 2023, № 4 86
cells of the root cortex are shed, and plants that were severely harmed earlier in
the season may not continue to exhibit symptoms later in the season.
Diseases on Leaves of Cotton.
The diseases on the cotton leaves are classified as:
– Bacterial disease: e.g. Bacterial Blight, Crown Gall, Lint Degradation;
– Fungal diseases: e.g. Anthracnose, Leaf Spot;
– Viral disease: e.g. Leaf Curl, Leaf Crumple, Leaf Roll;
Diseases like alternaria leaf spot, Bacterial blight, Bacterial stunt, Black
root, Boll rot/tight lock.
The collection of observations is divided into smaller groups so that observa-
tions within each group are somewhat comparable to one another. Multivariate
data analysis typically uses clustering as a routine practice. It is intended to inves-
tigate the data objects’ innate natural structure, where items in the same cluster
are as similar as possible to one another and objects in separate clusters are as dis-
tinct as possible from one another. The method used to arrange items or patterns
so that samples from the same group resemble one another more than samples
from other groups. There have been many different clustering techniques em-
ployed, including the hard clustering scheme and the fuzzy clustering scheme,
each of which has unique particular traits. Each data point can only belong to one
cluster when using the traditional hard clustering approach. As a result, when us-
ing this method, the segmentation results are frequently quite precise, meaning
that every pixel in the image belongs to exactly one class. Yet, in many actual
scenarios, problems with pictures like inadequate contrast, noise, overlapping in-
tensities, and insufficient spatial resolution make this hard (crisp) segmentation a
challenging process.
Types of Clustering:
Hard: same object can only belong to single cluster.
Soft: same object can belong to different clusters.
The current days, deep learning approach is used for cotton leaf segmenta-
tion. The CNN, VGG-16, VGG-19, ResNet-50 and some hybrid model has been
used for this problem. The deep learning approach has been improved the accu-
racy of cotton leaf image segmentation as compared to state-of-art. The
NMKFCM model is also gives stable result as compared to deep learning ap-
proaches. In deep learning approaches, training time period is major constraint
for this problem. In the experimental analysis, the training time is near about 1
hour to 2 hour and in the proposed method the training process is not required.
MATERIAL AND METHODS
Fuzzy Clustering
In image segmentation, a soft segmentation technique has received extensive
study and effective application. Since it has resilient qualities for ambiguity and
can preserve significantly more information than hard segmentation methods, the
Fuzzy C-Means (FCM) algorithm is the most widely used fuzzy clustering ap-
proach in picture segmentation. The typical FCM method has a severe flaw in that
Novel modified kernel fuzzy c-means algorithm used for cotton leaf spot detection
Системні дослідження та інформаційні технології, 2023, № 4 87
it lacks spatial context information, making it vulnerable to noise and imaging
artefacts even while it performs well on the majority of noise-free pictures.
Fuzzy C-Means Algorithm. A well-liked and practical image division
algorithm is FCM. The FCM algorithm was created by Dunn and enhanced
by Bezdek [1]. This method is intended to scale back an objective goal [2].
Because each quality vector may only belong to one cluster and the quality
vectors of the data set can be separated into solid clusters, this method out-
performs the k-mean technique. Instead, the FCM loosens the restriction
and enables the quality vector to assign a range of association scores to di-
verse clusters. Suppose a set of data with related clusters. A data value is
equidistant from both clusters while also being near to them.
Activity in the clustering loop is FCM. By reducing the intragroup biased
sum of the squared error task mJ function, it produces the best c partitions [3]:
2
,,
21
ji
m
ji
N
i
C
j
m dUJ
,
where N — the number of patterns in X ; C — the number of clusters; ijU —
the degree of membership; jW — the center of cluster j ; ijd — distance between
object iX and cluster center jW ; m — the biased value.
The FCM algorithm focuses on minimizing mJ , subject to the following
constraints on U :
,,,3,2,1,]1,0[ NiUij and ;,,3,2,1 Cj
C
j
ijU
1
,1 ,,,3,2,1 Ni
N
i
ijU
1
,10 .,,3,2,1 Cj
Objective function mJ describe a constrained optimization problem, which
can be converted to an unconstrained optimization problem by using Lagrange
multiplier technique. By using this calculates membership function and update
cluster center separately:
1
2
1
1
m
il
ijc
i
ij
d
d
U , Ni ,,2,1 , and Cj ,,2,1 ;
If 0ijd then 1ijU and 0ijU for j1 .
And calculate cluster center using following step
m
ij
N
i
i
m
ij
N
i
j
U
xU
w
)(
)(
1
1
, Cj ,,2,1 .
The FCM algorithm focuses on minimizing objective function Jm. It fails in
noisy image to detect accurate and sharp image segmentation process.
Kernel Fuzzy C-Means Algorithm. The FCM algorithm calculates the dis-
tance between the cluster center and the data item using Euclidian distance. FCM
Pradip M. Paithane, Sarita Jibhau Wagh
ISSN 1681–6048 System Research & Information Technologies, 2023, № 4 88
fails in noisy and nonlinear data sets because Euclidian distance does not perform
as intended in noisy data. Kernel Fuzzy C-means technique is used to address this
flaw. Kernel information is used with FCM in the KFCM approach [4]. KFCM
works by mapping input data into a higher-dimensional feature space and utilizing
the Kernel technique to transform nonlinearly separable data into linearly separa-
ble data. While using the kernel approach, the data set was complicated and
nonlinear before becoming simple and separable when using the FCM method [5].
KFCM classifies noisy objects into clusters with greater clarity than FCM
and with greater accuracy in noisy images. The value of the KFCM membership
matrix U may range from 0 to 1
KFCM is iterative clustering methods that generate optimal c partition by us-
ing minimize objective function kfcmJ :
)),(1(2),(
11
ji
m
ij
N
i
C
j
km WXKUWUJ
.
In this objective function Gaussian kernel function is used:
2
2
exp),(
yx
YXK .
In KFCM clustering algorithm choose initial cluster randomly and perform
following step.
1. Provide Gaussian kernel function for input image.
2. Evaluate membership function between object and cluster center.
3. Evaluate new updated cluster center.
4. Repeat step iteratively until no new cluster found.
KFCM it work properly in noisy image but KFCM not focus on neighbor-
hood term.
Modified Kernel Fuzzy C-Means (MKFCM). This method is intended to
scale back an objective goal [6]. Because each quality vector may only belong to
one cluster and the quality vectors of the data set can be separated into solid clus-
ters, this method outperforms the k-mean technique. Instead, the Fuzzy C-Mean
loosens the restriction and enables the quality vector to assign a range of
association scores to diverse clusters. Suppose a set of data with related clusters.
A data value is equidistant from both clusters while also being near to them.
FCM is looping clustering activity. It generates optimum c partitions by
abating the intragroup biased sum of the squared error task mJ function [7].
Kernel Method. The kernel methodology is a method that, by replacing the
internal product with an appropriate Mercer Kernel, generates an implicit non-line
map of the feedback information to a high-level quality space [8]. The kernel may
be used in any method that solely depends on the dot product between two vec-
tors. Every time a kernel is applied, a dot product is replaced. When two data are
planned into a high-level-dimensional space, the space metrical that calculates the
space between them is simplified. It is easier to tell apart and more distinctly dif-
ferentiated [9].
Novel modified kernel fuzzy c-means algorithm used for cotton leaf spot detection
Системні дослідження та інформаційні технології, 2023, № 4 89
Feature Space Mapping. Consider a non-line map task 32 : FI
from the 2-dimensional input space I into the 3-dimensional feature space F:
Txxxxx ) ,2,()( 2
221
2
1
. (1)
Hyperplane is represented by Eq. for separable dataset:
0 bxwT . (2)
Consider the splitting hyperplane Eq. (1) into a linear task in 3 :
02)( 2
23212
2
11 xwxxwxwxwT
.
Eq. (2) is an elliptic job when as usual value of a constant c and assessed in 2 .
In Fig. 1 any nonlinear separable data is converted into linear separable, so
every pixel is classified on the basis of a feature.
Use the appropriate mapping function to use F’s linear classifier with the
converted form of the data to find a non-straight classifier without hassle. After
mapping the non-line distinguishable data to a high-level space, I, locate a hyper-
plane that distinguishes linearly. For sensitive learning consider Fig. 1.
It depends only on the data mapped by the inner product of the feature space
F. Defining a function )()(),( xxxxK T
ii
, called kernel, that directly calcu-
lates the dot product of the mapping data places in the quality space eliminates the
need for even the explicit coordinates of F or the mapping task [10]. The subse-
quent standard sample of a kernel “K” shows the computation of the dot product
in the quality space applying 2),(),( ZXZXK T
. It is encouraging the map task
Txxxxx ),,2,()(Φ 2
221
2
1
:
),( 21 xxx , ),( 21 zzz ;
2),)(,( ZXZXK T
2
2211 )( zxzx
)2( 2
2
2
22211
2
1
2
1 zxzxzxzx ),2,(),2,( 2
221
2
1
2
221
2
1 zzzzxxxx T .
The advantage of such a kernel operation is that the complexness of the im-
provement of drawback continues solely reliant on the spatial property of the “in-
put space” and not of the “quality space”.
Fig. 1. Conversion of Non-line Distinguishable Data into Line Distinguishable Data
Pradip M. Paithane, Sarita Jibhau Wagh
ISSN 1681–6048 System Research & Information Technologies, 2023, № 4 90
Different types of Kernels are mentioned below [11]:
Linear Kernel function: zxzxK T),( ;
Polynomial Kernel function: dT zxzxK )(),( ;
Gaussian Kernel function:
2
2
exp),(
zx
zxK ;
Sigmoid Kernel function: )))(((tanh),( zxzxK T .
Figure 2 is depicted the kernel working process for liner separable process to
detect correct segmented regions. NMKFCM method is integrating closer pixel
quantity in objective function [12]. NMKFCM method is a revised form of
KFCM. KFCM is unsuitable for images damaged by instinct disturbance. KFCM
has operated accurately in indistinct and nonlinear separable data, but it doesn’t
consist of information of closer pixel, to overcome this drawback, introduced
NMKFCM is integrating closer pixel cost by applying “3×3” or “5×5” window
window. A closer pixel quantity is included in objective task [13; 14]. Thec “ ”
constraint is applied to manage the impact of closer’s term. It is having upper cost
with growth of image disturbance. Scale of cost rests within “0 to 1”, if ratio of
disturbance is minimal then take cost of between “0 and 0.5”. Ratio of distur-
bance is above average then take cost of is “0.5 and 1.0”. It is a beneficial and
useful algorithm as compared to other algorithms. It has achieved sharp outcomes
in disturbance images.
It is a looping procedure. It reduces the cost of objective tasks through closer
pixel. In this objective task, present window across pixel and “ ”parameter [15]:
Q
y
P
x R
Nk
ykR
yxT
m
xyNMKFCM N
UN
UJ i
obj
1 1
))W,(ZK1(W)(U, , (3)
where RN — the cardinality; iN — set of closer pixel value include into a win-
dow across pixel iZ . Objective task nmknj illustrate a constrained optimization
dilemma. Eq. 3 is applied for conversion into an unconstrained optimization di-
lemma. In Eq. 3, Lagrange multiplier technique is used.
Input Space Kernel Feature Space
Fig. 2. Kernel Feature Space
Novel modified kernel fuzzy c-means algorithm used for cotton leaf spot detection
Системні дослідження та інформаційні технології, 2023, № 4 91
By applying this computes membership function and update cluster center
separately:
1
1
1
1
))W,(ZK1(
))W,(ZK1(
m
R
Nl
klR
yxT
m
R
Nl
ylR
yxT
xy
N
UN
N
UN
U
l
i
.
And calculate cluster center using following step:
1
1
(Z , W ) Z
(Z , W )
P
m
xy T x y x
y
y Q
m
xy x y
x
U K
W
U
=
=
=
å
å
.
Algorithm: Objective Function of NMKFCM
INPUT
1. },,,{ 21 NZZZZ , Data set
2. yPP 2, , y is number of cluster
3. Define cost of Ɛ, used to terminate loop
4. Set membership function 0
xyU using input data and cluster.
5. Determine cluster center ),,,( 002010 pwwwW
OUTPUT
},,,{ 20 pj WWWW , targeted center of clusters.
begin
for
t=0
if }{ 1tt UU
Update center t
pW with tU by using Eq.
Update membership function 1tU by using Eq.
t+1
else
segmented output
end
This method is advantageous to integrate closer pixel information. Standard
FCM and IFCM methods are responsive to disturbance and preliminary cluster
centers. It is ignoring the 3-D correlation of pixels, leading to inaccurate cluster-
ing outcomes [16]. NMKFCM work very fit in neighborhood pixel material.
Goal and Objectives:
Choosing value of alpha to improve accuracy of image segmentation.
Pradip M. Paithane, Sarita Jibhau Wagh
ISSN 1681–6048 System Research & Information Technologies, 2023, № 4 92
Add Gaussian kernel method and RBF function to give more accuracy
and also work in noisy and noiseless.
Determine required number of clusters for image segmentation.
Improving CAR value in all image formats.
Hills Climbing Algorithm
Image segmentation is a crucial step in the processing of images. Applications
like image segmentation, adaptive compression, and region-based image retrieval
benefit from the detection of conspicuous picture areas. Saliency is measured by
comparing an image region’s local contrast to its surrounding area at different
scales [17]. It is using a contrast determination filter that runs at various scales to
produce saliency maps with saliency values per pixel for the purpose of identify-
ing salient locations. These separate maps come together to form the final sali-
ency map [18]. We employ a rather straightforward segmentation approach to
show how the final saliency map may be used to segment whole objects [19].
In this, compare the distance between the average feature vectors of the
pixels in a subregion of the picture with the pixels in the area around it. Instead of
merging separate saliency maps for scalar values of each feature, this enables the
creation of a combined feature map at a particular scale utilizing feature vectors
for each pixel [20]. The distance D between the average vectors of pixel charac-
teristics of the inner area 1R and that of the outer region 2R is what determines
the contrast-based saliency value ),( jic for a pixel at location )( jI in the picture:
q
N
q
p
N
p
ji v
N
v
N
Dc
21
1211
,
1
,
1
,
where v is the vector of feature elements corresponding to a pixel and 1N and 2N
are the number of pixels in 1R and 2R , respectively. If v is a vector containing
uncorrelated feature items, then the distance D is a Euclidean distance; if the vec-
tor’s elements are correlated, then the distance D is a Mahalanobis distance. In
this study, feature vectors for color and brightness are generated using the CIELab
color space and RGB photographs. Although the CIELab colour space’s percep-
tual differences are roughly Euclidian, D in equation [13]:
21, vvc ji ,
R1
R2
R2
Fig. 3. Saliency map with R1 inner and R2 outer region
Novel modified kernel fuzzy c-means algorithm used for cotton leaf spot detection
Системні дослідження та інформаційні технології, 2023, № 4 93
where TbaLv ];;[ 1111 and TbaLv ];;[ 2222 are the average vectors for regions
1R and 2R , respectively.
The final saliency map is determined as a sum of saliency values across the
scales S :
ji
s
ji cm ,, .
The hill-climbing technique may be thought of as a search window that is
ran through the d-dimensional histogram’s space to locate the biggest bin inside
of it. Each bin in the colour histogram has 2613 d neighbors because the
CIELab feature space is three-dimensional, where d is the number of dimensions
in the feature space. The values of these bins serve as the starting seeds, and the
number of peaks obtained reveals the value of K [21]. By adding up values in the
final saliency map M that correspond to pixels in the segmented picture, the av-
erage saliency value V per segmented region is determined:
ji
rjik
k m
r
V
k
,
,
1
,
kr is the segmented region’s size in pixels. The segments with an average sali-
ency value greater than a predetermined threshold T are maintained, while the
other segments are removed, according to a straightforward threshold-based pro-
cedure. As a consequence, the output only includes the segments that make up the
salient item.
The *** baL color space enables us to quantify these differences. The
*** baL color space is derived from the CIE XYZ tristimulus values. The
*** baL space comprises of a luminosity layer ‘ *L ’, chromaticity-layer ‘ *a ’
indicating where color falls along the red-green axis, and chromaticity-layer ‘ *b ’
indicating where the color falls along the blue-yellow axis [22].
Algorithm: Hill-climbing Based Segmentation.
Input: An Image.
Output: a group of aesthetically connected segments.
1. Create the image’s color histogram.
2. Ascend the color histogram’s slope from a non-zero bin to the apex as
shown below:
2.1. The amount of pixels in the current histogram bin should be com-
pared to the numbers in the adjacent (left and right) bins.
2.2. The algorithm moves upwards towards the neighboring bin with the
greater number of pixels if the surrounding bins have differing amounts of pixels.
2.3. The algorithm checks the next nearby bins if the immediate
neighbors have the same amount of pixels, and so on, until two neighboring bins
with different numbers of pixels are discovered. Next, a shift upward is performed
to the bin with the most pixels.
2.4. Repeat steps 2.1–2.3 to continue going upwards until you reach a
point from which you can travel no further uphill. When the adjacent bins contain
less pixels than the current bin, that is the situation. As a result, the present bin is
considered a high.
Pradip M. Paithane, Sarita Jibhau Wagh
ISSN 1681–6048 System Research & Information Technologies, 2023, № 4 94
Choose a different unclimbed bin to use as your starting bin, then follow step
2 to locate another summit. This process is repeated until the color histogram’s
non-zero bins are all climbed (associated with a peak). The discovered peaks are
preserved since they indicate the input image’s original number of clusters.
2.5. The halting bin is designated as the peak of a hill if no upward pro-
gress is made, and all bins going to this peak are connected to it.
3. Choose a different unclimbed bin to use as your starting bin, then follow
step 2 to locate another summit. This process is repeated until the histogram’s
non-zero bins are all climbed (associated with a peak).
4. The recognized peaks are preserved because they show how many clusters
there were in the input picture at the beginning.
5. The same peak’s neighboring pixels are clustered together.
Lastly, pixels that are close to one another and lead to the same peak are
grouped together, assigning each pixel to a different peak. Hence, create the input
image’s clusters.
EXPERIMENTAL RESULT
Evaluation Parameter
1. Cluster Accuracy Rate
A S
CAR
A S
.
2. Dice
( , ) 2
A S
dice A S
A S
.
3. IOU
A S
IOU
A S
.
4. Bfscore
2
( )
precision recall
bfscore
recall precision
,
where A output image; S input image.
Detail comparison of proposed method with traditional method (see Table 1–3)
T a b l e 1 . Detail Comparison of Proposed Method with Traditional Method
Evaluation Parameter
Image Name Approach
IOU bfscore dice
FCM 55.78 33.51 71.62
KFCM 73.81 26.36 84.93 Image 1
NMKFCM 81.55 41.25 89.83
FCM 68.61 21.28 81.38
KFCM 84.47 36.81 91.58 Image 2
NMKFCM 89.88 42.29 98.81
Novel modified kernel fuzzy c-means algorithm used for cotton leaf spot detection
Системні дослідження та інформаційні технології, 2023, № 4 95
Continued Table 1
Evaluation Parameter
Image Name Approach
IOU bfscore dice
FCM 90.22 37.14 94.86
KFCM 80.66 34.71 89.29 Image 3
NMKFCM 90.75 36.98 95.15
FCM 72.94 18.11 84.35
KFCM 90.06 40.33 94.77 Image 4
NMKFCM 94.51 54.46 97.18
FCM 74.95 26.51 85.68
KFCM 73.17 25.37 84.51 Image 5
NMKFCM 80.36 27.42 87.99
T a b l e 2 . Detail Comparison of Proposed Method with Traditional Method
Image Name Approach Cluster Accuracy Rate(CAR)
FCM 63.71
KFCM 71.78 Image 1
NMKFCM 74.98
FCM 57.9021
KFCM 64.8723 Image 2
NMKFCM 69.9572
FCM 67.2396
KFCM 64.3482 Image 3
NMKFCM 70.2246
FCM 58.6222
KFCM 66.3623 Image 4
NMKFCM 69.184
FCM 86.8277
KFCM 86.0188 Image 5
NMKFCM 95.337
T a b l e 3 . Detail comparison of proposed method with traditional method
Image Name Approach Time Period
FCM 12.24
KFCM 10.44 Image 1
NMKFCM 8.47
FCM 14.24
KFCM 08.37 Image 2
NMKFCM 06.54
FCM 11.61
KFCM 11.29 Image 3
NMKFCM 09.24
FCM 13.19
KFCM 12.66 Image 4
NMKFCM 07.29
FCM 12.24
KFCM 08.59 Image 5
NMKFCM 07.76
Pradip M. Paithane, Sarita Jibhau Wagh
ISSN 1681–6048 System Research & Information Technologies, 2023, № 4 96
The Table 4 depicts the detail comparison of NMKFCM approach with
deep learning approaches. The NMKFCM is providing strong and stable result
as compare to CNN model. The CAR value of NMKFCM approach is 98.80
which higher than other approaches. The IOU and Precision value of
NMKFCM achieved higher result value as compared to deep learning models.
The NMKFCM is having less value The Dice value of NMKFCM improvised
the result as compare to state-of-art. in bfscore as compare to other ap-
proaches.
T a b l e 4 . Detail comparison of proposed method with traditional method
Approach CAR IOU Precision Dice bfscore Time
Training
Time
CNN [23] 95.37 92.00 0.8750 46.0 87.50 8~9 second 94 Minute
VGG16 [23] 98.10 92.18 0.9583 46.0 95.16 5~6 second 54 Minute
ResNet-50 [23] 98.32 91.49 0.9482 50.0 95.65 2~3 second 53 Minute
Menon Model[23] 98.53 94.23 0.9579 50.0 96.42 5~6 second 77 Minute
NMKFCM 98.80 94.51 1.0000 98.81 54.46 4~6 second Not Required
The execution time of NMKFCM is less than CNN, VGG-16 and Menon
Model, but higher than ResNet-50. The training time is not required for
NMKFCM (Fig. 4).
In above image, sub image (A), (E), (I), (M) and (Q) are original image of
cotton leaf. Sub image (B), (F), (J), (N) and (R) are segmented by FCM approach,
Sub image (C), (G), (K), (O) and (S) are segmented by KFCM approach, Sub im-
age (D), (H), (L), (P) and (T) are segmented by NMKFCM approach. The sub
image (A) is affected by Bacterial Blight disease, The sub image (E) is affected
by Leaf Curl, The sub image (I) is affected by alternaria leaf spot , The sub image
(M) is affected by fungal disease.
0
20
40
60
80
100
120
CAR IOU Precision Dice bfscore
CNN
VGG‐16
ResNet‐50
Menon Model
NMKFCM
E
va
lu
at
io
n
P
ar
am
et
er
Evaluation Parameter
Fig. 4. Comparison of NMKFCM with Deep Learning Approaches
Novel modified kernel fuzzy c-means algorithm used for cotton leaf spot detection
Системні дослідження та інформаційні технології, 2023, № 4 97
CONCLUSION
The crucial and indispensable element of an image analysis system, image seg-
mentation is a key area of study for many image processing researchers. Four
methods — Clustering, Thresholding, Region Extraction, and Edge Detection —
are used to segment images. Clustering is the downgrouping of related data ele-
ments. Here, we’ve used techniques for clustering like crisp and fuzzy. In this
system, Fuzzy C-Means, Kernel Fuzzy C-Means, and Modified Kernel Fuzzy C-
Mean Clustering are all used as clustering techniques. In comparison to FCM and
Crips clustering methods, MKFCM is a suggested system that provides accurate
picture segmentation while also enhancing segmentation performance by adding
the influence of neighbor pixel information. The MKFCM method can automati-
cally identify the necessary cluster number for picture segmentation with the use
of the Hill climbing algorithm. The suggested technique can automatically esti-
mate the cluster number for a noisy picture, but this number is not helpful for im-
age segmentation since the proposed algorithm has formed a cluster for noisy pix-
els, making image segmentation less effective than for noiseless pixels. In the
future, we will be able to select an alpha value to increase the precision of picture
segmentation and CAR (Cluster Accuracy Rate) values across all image formats.
The proposed method is not required training time but in deep learning ap-
proaches training is mandatory. The proposed method is improvising the IOU,
precision, Dice and CAR value as compared to deep learning approaches.
No conflict of interest.
(A) (B) (C) (D)
(E) (F) (G) (H)
(I) (J) (K) (L)
(M (N) (O) (P)
(Q) (R) (S) (T)
Fig. 5. Cotton Leaf Image Segmentation using FCM, KFCM and NMKFCM
Pradip M. Paithane, Sarita Jibhau Wagh
ISSN 1681–6048 System Research & Information Technologies, 2023, № 4 98
REFERENCES
1. P.M. Paithane and S.N. Kakarwal, “Automatic determination number of cluster for multi
kernel NMKFCM algorithm on image segmentation,” in International Conference on
Intelligent Systems Design and Applications, pp. 870–879. Springer, Cham, 2018.
2. Chun-Yan Yu, Ying Li, Ai-lian Liu, and Jing-hong Liu, “A novel modified kernel
fuzzy c- means Clustering algorithms on Image segmentation,” 2011 14th IEEE In-
ternational Conference. doi: 10.1109/CSE.2011.109.
3. Saiful Islam and Dr. Majidul Ahmed, “Implementation of Image Segmentation for
Natural Images using Clustering Methods,” IJETAE, vol. 3, issue 3, March 2013.
4. L.A. Zadeh, “Fuzzy Sets”, Information and Control, 8, pp. 338–353, 1965.
5. Songcan Chan and Daoqiang Zhang, “Robust Image Segmentation Using FCM With
Spatial Constraints Based on New Kernel Induced Distance Measure,” IEEE trans-
actions on Systems, MAN and Cybernetics-Part B: Cybernetics, vol. 34, no. 4, Au-
gust 2004.
6. P.M. Paithane, S.N. Kakarwal and D.V. Kurmude, “Automatic Seeded Region
Growing with Level Set Technique Used for Segmentation of Pancreas,”
in International Conference on Soft Computing and Pattern Recognition, pp. 374–
382. Springer, Cham, 2020.
7. P.M. Paithane and S.N. Kakarwal, “Automatic Pancreas Segmentation using A Novel
Modified Semantic Deep Learning Bottom-Up Approach,” International Journal of
Intelligent Systems and Applications in Engineering, 10(1), pp. 98–104, 2022.
8. Pradip Mukundrao Paithane, “Yoga Posture Detection Using Machine Learn-
ing,” Artificial Intelligence in Information and Communication Technologies,
Healthcare and Education: A Roadmap Ahead, 2022.
9. S. Kakarwal and Pradip Paithane, “Automatic pancreas segmentation using ResNet-
18 deep learning approach,” System Research and Information Technologies, no. 2,
pp.104–116, 2022.
10. Elnomery Zanaty and Sultan Aljahdali,“Improving Fuzzy Algorithms for Automatic
Magnetic Resonance Image Segmentation,” The International Arab Journal of In-
formation Technology, vol. 7, no. 3, pp. 271–279, July 2009.
11. Kaur Prabhjot, Gupta Pallavi, and Sharma Poonam, “Review and Comparison of
kernel Based Fuzzy Image Segmentation Techniques,” I.J. Intelligent Systems and
Applications, 7, pp. 50–60, 2012.
12. Robert L. Cannon, Jintendra V. Dave, and James C. Bezdek, “Efficient Implementa-
tion of the Fuzzy c-Means Clustering Algorithms,” IEEE Transactions on Pattern
Analysis and Machine Intelligence, vol. PAMI-8, issue 2, March 1986.
13. Martin Hofman, Support vector Machines-Kernel and the Kernel Trick, pp. 1–16, 2006.
14. Daoqiang Zang and Songcang Chen, “Fuzzy Clustering Using Kernel Method,” Pro-
ceedings of the 2002 International Conference on Control and Automation, Xiemen,
China, June 2002. doi: 10.1109/ICCA.2002.1229535.
15. Daoqiang Zang and Songcang Chen, “A novel kernalized fuzzy C-means algorithm
with Application in medical image segmentation,” Artificial Intelligence in Medi-
cine, 32, pp. 37–50, 2004.
16. E.A. Zanaty, Sultan Aljahdli, and Narayan Debnath, “A Kernalized Fuzzy C-Means
Algorithm for Automatic Magnetic Resonance Image Segmentation,” Journal of
Computational Methods in Sciences and Engineering Archive, vol. 9(1,2S2),
pp. 123–136, April 2009. doi: 10.3233/JCM-2009-0241.
17. Shailash Kochra and Sanjay Joshi, “Study on Hill-Climbing Algorithm for Image
Segmentation Technology,” International Journal of Engineering Research and Ap-
plications (IJERA), vol. 2, issue 3, pp. 2171–2174, May-Jun 2012.
18. Garima Goyal, “TEM Color Image Segmentation using Hill Climbing Algorithm,”
International Journal of Computer Science and Information Technologies, vol. 5, pp.
3457–3459, 2014.
19. A. Abirami Shri, E. Aruna, and Ajanthaa Lakkshmanan,“Image segmentation and
recognition,” International Journal of Computer Applications; 3rd National Confer-
ence on Future Computing, February 2014.
Novel modified kernel fuzzy c-means algorithm used for cotton leaf spot detection
Системні дослідження та інформаційні технології, 2023, № 4 99
20. Pradip M. Paithane, S.N. Kakarwal, and D.V. Kurmude, “Top-down method used for
pancreas segmentation,” Int. J. Innov. Exploring Eng. (IJITEE), vol. 9, issue 3, pp.
1790–1793, 2020.
21. Sarita Jibhau Wagh, Pradip M. Paithane, and S.N. Patil, “Applications of Fuzzy
Logic in Assessment of Groundwater Quality Index from Jafrabad Taluka of Mara-
thawada Region of Maharashtra State: A GIS Based Approach,” Hybrid Intelligent
Systems: 21st International Conference on Hybrid Intelligent Systems (HIS 2021),
December 14–16, 2021. Cham: Springer International Publishing, 2022.
22. Pradip M. Paithane and Sarita Jibhau Wagh, “Automatic Quality Control Scrutiny of Sugar
Crystal using K-Means Clustering Algorithm Image Processing,” American Scientific
Research Journal for Engineering, Technology, and Sciences, 9(12):2395-0056, 2022.
23. M.S. Memon, P. Kumar, and R. Iqbal, “Meta Deep Learn Leaf Disease Identification
Model for Cotton Crop,” Computers, 11(7), 102, 2022. Available:
https://doi.org/10.3390/computers11070102
24. Pradip Paithane, Sarita Jibhau Wagh, and Sangeeta Kakarwal, “Optimization of
route distance using k-NN algorithm for on-demand food delivery,” System Research
and Information Technologies, no. 1, pp. 85–101, 2023. doi: 10.20535/SRIT.2308-
8893.2023.1.07.
Received 01.04.2023
INFORMATION ON THE ARTICLE
Pradip M. Paithane, ORCID: 0000-0002-4473-7544, Vidya Pratishthan’s Kamalnayan
Bajaj Institute of Engineering and Technology, India, e-mail: paithanepradip@gmail.com
Sarita Jibhau Wagh, ORCID: 0000-0003-4798-2147, T.C. College Baramati, India
НОВИЙ МОДИФІКОВАНИЙ АЛГОРИТМ ЯДРА FUZZY C-MEANS, ЩО
ВИКОРИСТОВУЄТЬСЯ ДЛЯ ВИЯВЛЕННЯ ПЛЯМ НА ЛИСТУ
БАВОВНИКА / Прадіп М. Пейтане, Саріта Джібхау Ваг
Анотація. Сегментація зображення є важливою та складною темою, яка є не-
обхідною умовою як для базового аналізу зображення, так і для складної ін-
терпретації зображення. В аналізі зображень сегментація зображення має ви-
рішальне значення. Кілька різних програм, зокрема ті, що стосуються
медицини, ідентифікації обличчя, діагностики хвороби Коттона та виявлення
об’єктів на карті, отримують переваги від сегментації зображення. Для сегмен-
тації зображень використовується підхід кластеризації. Існує два типи алгори-
тмів кластеризації: чіткий і нечіткий. Техніка чіткості перевершує нечітку кла-
стеризацію. Нечітка кластеризація використовує добре відомий підхід FCM
для поліпшення результатів сегментації зображення. Техніка KFCM для сег-
ментації зображення може бути використана для усунення недоліків FCM у
зашумлених і нелінійних роздільних зображеннях. У підході KFCM ядрова
функція Гауса використовується для перетворення високовимірних нелінійно
розділених даних у лінійно розділені дані перед застосуванням FCM до даних.
KFCM поліпшує результати сегментації зображення із шумом, підвищує рі-
вень точності, але ігнорує сусідні піксели. Щоб подолати цю проблему, вико-
ристовується модифікований підхід нечіткого С-середнього ядра. Підхід
NMKFCM поліпшує результати сегментації зображення шляхом включення
інформації про сусідні піксели до цільової функції. Цей запропонований метод
використовується для виявлення плям «чорної шкірки» на листу бавовника.
Грибкове захворювання листя під назвою «чорна плямистість» призводить до
коричневого листя з фіолетовими краями. Бактерія може завдати шкоди рос-
линам бавовника, спричиняючи кутасті плями на листу, які мають колір від
червоного до коричневого.
Ключові слова: коефіцієнт точності кластера (CAR), кластеризація, хвороба
листя бавовника, метод нечіткої кластеризації (FCM), алгоритм нечіткого
C-середнього ядра (KFCM), новий модифікований алгоритм кластеризації не-
чіткого C-середнього ядра (NMKFCM).
|
| id | journaliasakpiua-article-297405 |
| institution | System research and information technologies |
| keywords_txt_mv | keywords |
| language | English |
| last_indexed | 2025-07-17T10:28:26Z |
| publishDate | 2023 |
| publisher | The National Technical University of Ukraine "Igor Sikorsky Kyiv Polytechnic Institute" |
| record_format | ojs |
| resource_txt_mv | journaliasakpiua/85/f9e6e0f4eee7cdefb449fbc963d06285.pdf |
| spelling | journaliasakpiua-article-2974052024-02-01T21:03:07Z Novel modified kernel fuzzy c-means algorithm used for cotton leaf spot detection Новий модифікований алгоритм ядра fuzzy c-means, що використовується для виявлення плям на листу бавовника Paithane, Pradip Wagh, Sarita Jibhau коефіцієнт точності кластера (CAR) кластеризація хвороба листя бавовника метод нечіткої кластеризації (FCM) алгоритм нечіткого C-середнього ядра (KFCM) новий модифікований алгоритм кластеризації нечіткого C-середнього ядра (NMKFCM) Cluster Accuracy Rate (CAR) Clustering Cotton Leaf Disease Fuzzy Clustering Method (FCM) Kernel Fuzzy C-means Algorithm (KFCM) Novel Modified Kernel Fuzzy C-Means Clustering Algorithm (NMKFCM) Image segmentation is a significant and difficult subject that is a prerequisite for both basic image analysis and sophisticated picture interpretation. In image analysis, picture segmentation is crucial. Several different applications, including those related to medicine, facial identification, Cotton disease diagnosis, and map object detection, benefit from image segmentation. In order to segment images, the clustering approach is used. The two types of clustering algorithms are Crisp and Fuzzy. Crisp clustering is superior to fuzzy clustering. Fuzzy clustering uses the well-known FCM approach to enhance the results of picture segmentation. KFCM technique for image segmentation can be utilized to overcome FCM’s shortcomings in noisy and nonlinear separable images. In the KFCM approach, the Gaussian kernel function transforms high-dimensional, nonlinearly separable data into linearly separable data before applying FCM to the data. KFCM is enhancing noisy picture segmentation results. KFCM increases the accuracy rate but ignores neighboring pixels. The Modified Kernel Fuzzy C-Means approach is employed to get over this problem. The NMKFCM approach enhances picture segmentation results by including neighboring pixel information into the objective function. This suggested technique is used to find “blackarm” spots on cotton leaves. A fungal leaf disease called “blackarm” leaf spot results in brown leaves with purple borders. The bacterium can harm cotton plants, causing angular leaf blotches that range in color from red to brown. Сегментація зображення є важливою та складною темою, яка є необхідною умовою як для базового аналізу зображення, так і для складної інтерпретації зображення. В аналізі зображень сегментація зображення має вирішальне значення. Кілька різних програм, зокрема ті, що стосуються медицини, ідентифікації обличчя, діагностики хвороби Коттона та виявлення об’єктів на карті, отримують переваги від сегментації зображення. Для сегментації зображень використовується підхід кластеризації. Існує два типи алгоритмів кластеризації: чіткий і нечіткий. Техніка чіткості перевершує нечітку кластеризацію. Нечітка кластеризація використовує добре відомий підхід FCM для поліпшення результатів сегментації зображення. Техніка KFCM для сегментації зображення може бути використана для усунення недоліків FCM у зашумлених і нелінійних роздільних зображеннях. У підході KFCM ядрова функція Гауса використовується для перетворення високовимірних нелінійно розділених даних у лінійно розділені дані перед застосуванням FCM до даних. KFCM поліпшує результати сегментації зображення із шумом, підвищує рівень точності, але ігнорує сусідні піксели. Щоб подолати цю проблему, використовується модифікований підхід нечіткого С-середнього ядра. Підхід NMKFCM поліпшує результати сегментації зображення шляхом включення інформації про сусідні піксели до цільової функції. Цей запропонований метод використовується для виявлення плям "чорної шкірки" на листу бавовника. Грибкове захворювання листя під назвою "чорна плямистість" призводить до коричневого листя з фіолетовими краями. Бактерія може завдати шкоди рослинам бавовника, спричиняючи кутасті плями на листу, які мають колір від червоного до коричневого. The National Technical University of Ukraine "Igor Sikorsky Kyiv Polytechnic Institute" 2023-12-26 Article Article application/pdf https://journal.iasa.kpi.ua/article/view/297405 10.20535/SRIT.2308-8893.2023.4.07 System research and information technologies; No. 4 (2023); 85-99 Системные исследования и информационные технологии; № 4 (2023); 85-99 Системні дослідження та інформаційні технології; № 4 (2023); 85-99 2308-8893 1681-6048 en https://journal.iasa.kpi.ua/article/view/297405/290389 |
| spellingShingle | коефіцієнт точності кластера (CAR) кластеризація хвороба листя бавовника метод нечіткої кластеризації (FCM) алгоритм нечіткого C-середнього ядра (KFCM) новий модифікований алгоритм кластеризації нечіткого C-середнього ядра (NMKFCM) Paithane, Pradip Wagh, Sarita Jibhau Новий модифікований алгоритм ядра fuzzy c-means, що використовується для виявлення плям на листу бавовника |
| title | Новий модифікований алгоритм ядра fuzzy c-means, що використовується для виявлення плям на листу бавовника |
| title_alt | Novel modified kernel fuzzy c-means algorithm used for cotton leaf spot detection |
| title_full | Новий модифікований алгоритм ядра fuzzy c-means, що використовується для виявлення плям на листу бавовника |
| title_fullStr | Новий модифікований алгоритм ядра fuzzy c-means, що використовується для виявлення плям на листу бавовника |
| title_full_unstemmed | Новий модифікований алгоритм ядра fuzzy c-means, що використовується для виявлення плям на листу бавовника |
| title_short | Новий модифікований алгоритм ядра fuzzy c-means, що використовується для виявлення плям на листу бавовника |
| title_sort | новий модифікований алгоритм ядра fuzzy c-means, що використовується для виявлення плям на листу бавовника |
| topic | коефіцієнт точності кластера (CAR) кластеризація хвороба листя бавовника метод нечіткої кластеризації (FCM) алгоритм нечіткого C-середнього ядра (KFCM) новий модифікований алгоритм кластеризації нечіткого C-середнього ядра (NMKFCM) |
| topic_facet | коефіцієнт точності кластера (CAR) кластеризація хвороба листя бавовника метод нечіткої кластеризації (FCM) алгоритм нечіткого C-середнього ядра (KFCM) новий модифікований алгоритм кластеризації нечіткого C-середнього ядра (NMKFCM) Cluster Accuracy Rate (CAR) Clustering Cotton Leaf Disease Fuzzy Clustering Method (FCM) Kernel Fuzzy C-means Algorithm (KFCM) Novel Modified Kernel Fuzzy C-Means Clustering Algorithm (NMKFCM) |
| url | https://journal.iasa.kpi.ua/article/view/297405 |
| work_keys_str_mv | AT paithanepradip novelmodifiedkernelfuzzycmeansalgorithmusedforcottonleafspotdetection AT waghsaritajibhau novelmodifiedkernelfuzzycmeansalgorithmusedforcottonleafspotdetection AT paithanepradip novijmodifíkovanijalgoritmâdrafuzzycmeansŝovikoristovuêtʹsâdlâviâvlennâplâmnalistubavovnika AT waghsaritajibhau novijmodifíkovanijalgoritmâdrafuzzycmeansŝovikoristovuêtʹsâdlâviâvlennâplâmnalistubavovnika |