SPECIAL SECTION ON NEW TECHNOLOGIES FOR SMART FARMING 4.

0:
RESEARCH CHALLENGES AND OPPORTUNITIES

Received February 28, 2019, accepted March 21, 2019, date of publication March 28, 2019, date of current version April 17, 2019.
Digital Object Identifier 10.1109/ACCESS.2019.2908040

An Optimized Method for Segmentation and

Classification of Apple Diseases Based on Strong
Correlation and Genetic Algorithm Based
Feature Selection
MUHAMMAD ATTIQUE KHAN 1 , M IKRAM ULLAH LALI 2 , MUHAMMAD SHARIF 3,

KASHIF JAVED4 , KHURSHEED AURANGZEB 5 , SYED IRTAZA HAIDER 5 ,

ABDULAZIZ SAUD ALTAMRAH5 , AND TALHA AKRAM3
1 Department of Computer Science and Engineering, HITEC University, Taxila, Pakistan
2 Department of Computer Science, University of Gujrat, Gujrat 50700, Pakistan
3 Department of Computer Science, COMSATS University Islamabad, Wah Campus, Wah, Pakistan
4 Department of Robotics, SMME, Islamabad 44000, Pakistan
5 Computer Engineering Department, College of Computer and Information Sciences, King Saud University, Riyadh 11543, Saudi Arabia

Corresponding authors: M Ikram Ullah Lali (ikramullah@uog.edu.pk) and Khursheed Aurangzeb (kaurangzeb@ksu.edu.sa)
This work was supported by the Deanship of Scientific Research at King Saud University for funding this work through research group
under Grant RG-1438-034.

ABSTRACT Agriculture is a major part of the world economy as it provides food safety. However, in recent
years, it has been noted that plants are extensively infected by different diseases. This causes enormous
economic losses in agriculture industry around the world. The manual inspection of fruit diseases is a
difficult process which can be minimized by using automated methods for detection of plant diseases at the
earlier stage. In this paper, a new method is implemented for apple diseases identification and recognition.
Three pipeline procedures are followed by preprocessing, spot segmentation, and features extraction, and
classification. In the first step, the apple leaf spots are enhanced by a hybrid method which is the conjunction
of 3D box filtering, de-correlation, 3D-Gaussian filter, and 3D-median filter. After that, the lesion spots
are segmented by the strong correlation-based method and optimized their results by fusion of expectation
maximization (EM) segmentation. Finally, the color, color histogram, and local binary pattern (LBP) features
are fused by comparison-based parallel fusion. The extracted features are optimized by genetic algorithm
and classified by One-vs-All M-SVM. The experimental results are performed on plant village dataset. The
proposed methodology is tested for four types of apple disease classes including healthy leaves, Blackrot,
Rust, and Scab. The classification accuracy shows the improvement of our method on selected apple diseases.
Moreover, the good preprocessing step always produced prominent features which later achieved significant
classification accuracy.

INDEX TERMS Symptoms enhancement, symptoms segmentation, feature extraction, optimal features,
recognition.

I. INTRODUCTION diseases are judged by the naked eyes of an expert in this

Efficient identification and recognition of fruit leaf diseases
is a current challenge in computer vision (CV) due to their
important applications in agriculture and agro-economy [1].
In agriculture, various types of fruit diseases exist which
affect the production and quality of fruits. Most of these
The associate editor coordinating the review of this manuscript and
approving it for publication was el-Hadi M. Aggoune.
area on the basis of their symptoms. However, it is expen-
sive due to the unavailability of experts [3] and higher cost.
In this regards, the computing researchers in collaboration
with agriculture experts have suggested many algorithms for
automated detection of diseases in plants and fruits [2].
Leaf symptoms are an important source of information to
identify the diseases in several different types of fruit plants.
Apple is an important fruit plant and widely famous for its

2169-3536 2019 IEEE. Translations and content mining are permitted for academic research only.
VOLUME 7, 2019 Personal use is also permitted, but republication/redistribution requires IEEE permission. 46261
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
 M. A. Khan et al.: Optimized Method for Segmentation and Classification of Apple Diseases
nutrient values. However, its production and quality are dam-
aged by the attack of different diseases like black rot, rust,and
blight. Therefore, it is essential to develop an automated com-
puterized system for detection and classification of apple leaf
symptoms at early stage. The early detection of these symp-
toms are helpful to improve the fruits quality and production.
In the field of computer vision, it is an active research area
to find out the lesion spot [4] and recently several methods
are introduced for fruits diseases detection through image
processing and machine learning algorithms [5]. Normally,
the apple lesion spots are examined by their color, texture,
and shape, therefore, the features like color, texture, and
shape are essential for classification of apple diseases. A lot
of segmentation, features extraction, and classification tech-
niques are proposed in literature for fruits diseases segmenta-
tion and recognition such as combination of features based
symptoms identification [6], color based segmentation [7],
correlation based features selection [4], optimized weighted
segmentation [8], texture features [9], shape features [10],
Support Vector Machine (SVM) [11], and few more [12].
A. PROBLEM STATEMENT
Various global challenges exist in the agriculture domain
such as change of environment and appearance of multiple
symptoms in one image. The complex background and mul-
tiple symptoms regions are not easy to detect from an image.
The presence of symptom regions on the boundary of image,
various characteristics of leaf symptoms, the existence of
noise, change in lighting condition, and color based similarity
among various symptoms. Each lesion spot has a different
shape, color, and texture which makes the detection process
difficult which later utilized in the classification phase. The
challenges which researchers faced in this domain using com-
puter vision and machine learning techniques are: a) correctly
segmentation of leaf symptoms in the image which exist
in various symptoms and most are similar to background
regions, (b) extraction of sufficient number of features for
classification because the irrelevant features do not produce
good accuracy and also effects on the system execution time,
and (c) selection of most prominent features. The justification
behind these key challenges of any computer vision system
are: (a) the correct symptoms segmentation is performed only
through clear difference among symptoms pixels and healthy
pixels, which are defined through a good preprocessing step,
(b) a good computer-based automated method always needs
a minimum good number of features for sufficient accuracy
along better execution time.
B. CONTRIBUTIONS
A new automated apple leaf diseases method is implemented
for detection and recognition through symptoms segmenta-
tion and best features selection. Three pipeline procedures
including image enhancement, lesion spot segmentation,
and prominent features selection & classification are uti-
lized. Our major contributions are in contrast enhancement,
46262
TABLE 1. List of abbreviations and their description.
segmentation, and feature selection blocks, which are enu-
merated below.
a) A hybrid contrast stretching technique is proposed in
which a 3D-Box filter is applied, which initially blurs the
boundary of a lesion spot, followed by an implementation of a
de-correlation function to reduce image distortion. This com-
plete procedure is facilitated by 3D-Gaussian and 3D-Median
filter - making the foreground and background maximally
differentiable.
b) A diseased spot segmentation algorithm is proposed in
which ’S’ channel of HSI color space is considered, which is
later used to identify set of clusters having higher pixel values.
A strong correlation (SC) based method is later proposed to
select a strong cluster, followed by an optimization step using
expectation maximization (EM) algorithm. This step plays its
role especially under the scenario when the infected regions
are small in size.
c) A method is designed to utilize color and tex-
ture features, which are later fused through parallel
approach. An addition of feature selection step using
genetic algorithm (GA) further expedite the process for
getting improved classification results using support vector
machines.
C. ABBREVIATIONS AND ACRONYMS
II. RELATED WORK
The common apple plant diseases are apple rust, apple rot,
apple scab, apple greening, and powdery mildew. Recently,
several researchers introduced various techniques such as
Dubey and Jalal [13] described a new computer vision
based method using multiple features. The described method
followed three major steps- i) diseased regions are seg-
mented by K-Means, ii) color, LBP and shape features are
extracted using segmented lesion spots. Finally, classification
VOLUME 7, 2019
M. A. Khan et al.: Optimized Method for Segmentation and Classification of Apple Diseases
is performed through MSVM and experiments are done on
four types of apple diseases including apple blotch, rot,
scab, and healthy. For each disease class, 80 images are
collected from Google and achieved maximum classification
accuracy on different feature vectors 98.75%, 95%, 96.25%,
and 100%, respectively. Khan et al. [6] presented an auto-
mated system for multiple fruit diseases using deep learning.
The presented method process into two primary steps- dis-
ease segmentation through correlation based approach and
classification of fruit diseases by deep features extraction
from pre-trained CNN models like VGG16 and AlexNet.
Chuanlei et al. [4] introduced a new image processing based
technique for the recognition of apple diseases. In the
presented method, RGB color transformation structure is
designed and segment the symptoms through region growing.
Finally, features are extracted for each spot with reduced
dimensions of extracted features by GA and correlation coef-
ficients. For experimental results, 90 diseased leaves images
are selected from three types of disease classes including
powdery mildew, mosaic, and rust with classification rate
is 90%. Dubey and Jalal [14] utilized a K-Means clustering
method for apple fruit diseases segmentation. After that,
extract texture (LBP, CLBP), color and global color his-
togram features and combined using simple concatenation
method. For experimental results, three types of apple fruit
disease images are selected and achieved recognition accu-
racy up to 93%. Also, the described method performs well for
CLBP features and achieved recognition accuracy 89.88%,
90.71%, 96.66%, and 99.33% for apple blotch, rot, scab, and
healthy. Ormani et al. [15] introduced an automated tech-
nique for detection of apple diseases. The introduced method
is based on three series of steps. In the first step, the lesion
spots are segmented by K-Means clustering method. Then,
the color, GLCM, and wavelet features are extracted from
segmented spots, which are classified by support vector
regression (SVR). The major advantage of SVR algorithm
is to minimize the error rate and increase the classification
accuracy. Shuaibu et al. [16] presented an optimized stochas-
tic algorithm for identification of apple marssonina disease.
A PSO algorithm is utilized to find out the optimal features
based on their highest discrimination. The selected features
are classified by SVM and achieved 100% accuracy. Sing
and Misra [17] described a two step automated technique.
The disease lesions spot are segmented by an optimized
genetic algorithm. The GA perform significant well with
SVM and achieved the segmentation accuracy 95.71%, which
is significantly better as compared to MDC with K-Means
and MDC with the proposed method. The major advantage
of GA is that it can process a large number of variables
at the same time. Barbedo [18] introduced a new method
for segmentation of disease spots on plant leaves based on
manipulation of a color channel and boolean operation, which
is performed on the binary mask. The introduced method
resolves many challenges such as variability and change in
condition. For experimental results, 77 disease images are
selected of 11 plants including citrus, coffee, corn, cotton, and
VOLUME 7, 2019
few more. Ali et al. [19] described an automated method for
detection and classification of citrus diseases. The described
method is based on two major steps including symptoms seg-
mentation and feature extraction. In segmentation, the Delta
E method is utilized and later on, texture and color features
are extracted. The extracted features are reduced by PCA
and fed to SVM for classification, which achieved maximum
recognition accuracy 99.9%.
III. PROPOSED METHODOLOGY
Three primary steps are incorporated int he proposed system-
contrast enhancement and symptom segmentation, multi-type
features extraction, and prominent features selection and
recognition, main architecture demonstrated in Figure. 1.
A. DISEASE SPOT ENHANCEMENT
Preprocessing step has gained much importance in CV and
image processing because of low contrast images effects
overall accuracy. This step is essential for later work on
disease spot segmentation and feature extraction because dis-
eases can be usually identified according to the symptoms
shape, color, and texture. In this article, Plant Village dataset
is used for evaluation. The Plant Village dataset contains
several types of problems such as noise, brightness, and
change of the origin. In this article, an optimized algorithm
is implemented to address these problems and to improve
the performance of quantitative image analysis. The pro-
posed method consists of three pipeline procedures including
3D-Box filtering, De-correlation, and the combination of a
3D-Gaussian and 3D-Median filter.
1) 3D-BOX FILTERING
For smoothing the disease spot region in the image,
3D-Box filter operation is performed. The 3D-box filter is
used equally weighting a rectangular neighborhood of pixels
in three major steps: a) compute filter size; b) padding, and c)
normalization factor. Let φxyz is an RGB input image having
dimension 256 × 256. Let ϕabc be an intensity value of size
a × b × c 3-dimensional image ϕ at position (m, n, o), while
ϕ is 3D-box filter into φ. Then, the size of the filter is defined
as follows:
a1 X
b1 X
c1
X
ϕijk = (ϕmno ) (1)
m=1 n=1 o=1
where, ϕijk denotes the size of filter, a1 = 4, b1 = 4, and
c1 = 3. After that, we perform padding on φxyz using filter
size as:
ϕijk − 1
PD = (2)
2
where, PD denotes the padding window, which is performed
on RGB image φxyz as follows:
N
X
Pd (φ) = (φxyz , PD ) (3)
x,y,z=1
46263
 M. A. Khan et al.: Optimized Method for Segmentation and Classification of Apple Diseases

FIGURE 1. Flow diagram of proposed detection and classification of apple diseases.

where, Pd (φ) is padding image. Each padding window is
updated horizontally based on adding a padding window on
the right side and subtracting out a value from the left side.
The updating of padding window is defined as follows:
P the mean of BF(x, y), and µt is mean of target output image
Pd (φ)
Update(Pd (φ)) = (4)
t as follows:
where t is a constant parameter for both left and right window
at the time of updating. Finally, performed normalization
on the image after completion of all padding iterations as
follows:
1 put T (x, y) to perform an eigen decomposition of covariance
NLφ = Q
(Pd (φ))
Q T (x, y) = σt × cov(ui , vj )σ −1
where, NLφ denotes the normalization factor, and ()
denotes the product of padding image. The effects of
3D-bx filtering operation on the original RGB image is shown
in Figure 2.
2) COLOR DIFFERENTIATION
Color differentiation is used to improve the color difference
between each pixel in the image. Therefore, to address this
problem, we used decorrelation. Decorrelation is a general
term for any process that is used to reduce the serial corre-
lation between two similar processes. It is done on the basis
of band to band correlation. Therefore, the excessive color
improves the visual perception of an image and also make it
easier for feature discrimination.
Let BF(x, y) denotes the 3D-box filter image, which is
obtained from equation 1 to 5. Let DC(x, y) denotes the target
output decorrelation image. Then, the vector x ∈ 8ui ×vj ,
where ui and vj denotes the nbands − by − 1 vector of 3D-
box filter image BF(x, y) and target output image DC(x, y).
Therefore, the vector ui containing the pixel value of BF(x, y)
is transformed into the corresponding pixel vj in the output
image DC(x, y) as follows:
vj = T (x, y) × (ui − µ) + µt (6)
where, T (x, y) is an nBands − by − nBands matrix, µ denotes
DC(x, y). Then, calculate the covariance matrix of ui and vj
Cov(ui , vj ) = E (ui − µi )(vj − µj )
 
(7)
Cov(ui , vj ) = E (ui vj )(µi µj )
 
(8)
where, µ = E(ui ) or E(vj ). After that,we compute target out-
(5) matrix as:
(9)
where, σt denotes the standard deviation (SD) of target out-
put, and σ −1 is SD of input
q 3D-box filter image. The SD (σ )
is computed as: σ = E(Cov(ui , vj ))2 − [E(Cov(ui , vj ))]2 .
Hence, the final enhanced decorrelation image is obtained as
follows:
DCF (x, y) = µt + σt × cor(ui , vj )σ −1 × (ui − µ) (10)
where, cor(ui , vj ) = σ × cov(ui , vj )σ −1 . The effects of
decorrelation stretching is shown in Figure 3.
3) IMAGE NORMALIZATION
In general, the image normalization is a process that improves
the range of pixel intensity values which is sometimes called
contrast stretching [21]. Normalization is performed in form
of smoothing the image and removing the noise factor. The
enhanced decorrelation image (DCF (x, y)) is used for this
processes because it contains too much noise as shown in
Figure 4. Therefore, we used 3D-Gaussian and 3D Median
filter to remove these artifacts. The Gaussian filter smoothes

46264 VOLUME 7, 2019

M. A. Khan et al.: Optimized Method for Segmentation and Classification of Apple Diseases

FIGURE 2. 3D-box filtering operation effects. a) Original image; b) 3D-box filter image, and c) histogram
of 3D-box filter image.

FIGURE 3. Color differentiation effects using decorrelation. a) 3D-box filter image; b) decorrelation image, and c) histogram of
decorrelation image.

the disease spot region but at the same time it reduces the process, the 3D Gaussian filter [20] is defined as:
disease information, therefore, an extra 3D Median filter is 2 +y2 +z2
performed after performing the 3D-Gaussian filter on the 1 −( x )
G33(x, y, z) = e 2π σ 2 (14)
3
image. σ 3 (2π) 2
Let DCF (x, y) denotes the three dimensional enhanced
decorrelation image having dimension 256 × 256 × 3. Let where, G33(x, y, z) is 3D-Gaussian filter image, whose
G(x, y) denotes the 2D gaussian filter, which is defined as effects are given in Figure 4(d). Finally, a 3D-Median filter
follows: (3 × 3 × 3) is performed on Gaussian filtered image, which
efficiently removes the remaining noise after performing 3D-
1 2 2 Gaussian filter. The effects of enhancement after performing
−( x +y )
G(x, y) = √ e 2σ 2 (11) 3D-Median filter image are shown in Figure 4(e).
σ 2π

where, σ 2 denotes the variance of given image and σ is SD. IV. DISEASE SEGMENTATION
The variance σ 2 and σ is computed as follows: Plant disease is one of the major factors which impacts to
diminish quantity and reduce the quality of the product.
P
(DCF (x, y))2
P
(DCF (x, y)) 2 Therefore, to control the plant diseases, it is much important
σ =
2
−( ) (12) to diagnose them at early stages. Several image process-
N N
s ing methods have been utilized for segmentation of such
(DCF (x, y))2
P P
(DCF (x, y)) 2 diseases in agriculture such as Fuzzy-C Means, K-Means
σ2 = −( ) (13)
N N Clustering, Otus segmentation, and color segmentation. From
all segmentation methods, the color segmentation is much
where, N denotes the total number of pixels in the input important because each disease spot has a different color.
DCF (x, y) image. To following above 2D gaussian filter Moreover, there are several challenges including the shape

VOLUME 7, 2019 46265

 M. A. Khan et al.: Optimized Method for Segmentation and Classification of Apple Diseases
FIGURE 4. Image Normalization effects. a) Original image; b) 3D-box filtering image; c) decorrelation image; d) 3D-Gaussian image,
and e) 3D-Median filtered image.
of disease spot, change in color and texture, and change in
origin and scale. To address these challenges, we present
an optimized disease segmentation method based on Strong
Correlated Pixels (SCP) and Expectation Maximization (EM)
algorithm. The optimization is performed based on the fusion
of SCP and EM by entropy method as shown in Figure 5.
The detailed description of the optimized method is given
below.
As G33(x, y, z) is an enhanced RGB image, where
(x, y, z) denotes its pixel values. Suppose, R(u, v) =
G33(x,y,z) , , G(u, v) = G33(x,y,z) , and B(u, v) = G33(x,y,z)
G33(red) G33(green) G33(blue)
denote the extracted red, green, and blue channel from
enhanced image. Then, HSI transformation [22] is performed
46266
and extract saturated channel (S) as as follows:
3 × min(R(u, v), G(u, v), B(u, v))
S(u, v) = 1 − (15)
G33(x, y, z)
where, S(u, v) is saturated channel and overall effects are
shown in Figure 5, which shows that the disease spots are
clearly highlighted after performing HSI transformation.
After that, we calculate the histogram of selected S channel
based on their mesh representation. Then, the number of bins
based on the higher brightness probability value is calculated.
The higher brightness value is computed based on their his-
togram as given in Figure 13(c). The probability is defined as
VOLUME 7, 2019
M. A. Khan et al.: Optimized Method for Segmentation and Classification of Apple Diseases

FIGURE 5. HSI transformation effects. a) 3D-Median filtered image; b) HSI

transformation; c) histogram representation of HSI transformation; d) selected S
channel; e) contour representation on selected S channel; f) mesh representation of S
channel.

follows:
pi
Prob(pix) = P (16)
(P(pix))
where, Prob(pix) denotes the probability value of image
pixels, pi denotes the intensity value of each pixel, and
P
(P(pix)) denotes the intensity value of all pixels in the FIGURE 6. Initial disease spot extraction effects. a) Selected S channel; b)
mesh representation; c) histogram of mesh representation; d) initial
image. Then, we select the lower intensity value probability binary image.
pixels in one cluster and later in other clusters. Let C1 denotes
the cluster which have higher probability intensity values (in where, ED(xi ) denotes the distance pixel values, which are
other words those pixels which pixels range is near to 180 or later utilized in threshold function to convert image into
greater), which means that those pixels which are ≥ 180. Let binary form.
C2 denotes those pixels, which pixel range is < 180. Then (
select the higher intensity value pixels as follows: 1 if xi ≤ 0.35
T (xi ) = (19)
( 0 Otherwise
C1 if Prob(bin) ≥ 180
SL(bin) = (17) The threshold function T (xi ) gives a binary image, which is
C2 if Prob(bin) < 180 shown in Figure 6. In Figure 13(a) the highlighted regions
are disease spots, (b) denotes the mesh representation of
where SL(bin) denotes the clustered matrix. Then, C1 is
highlighted regions, (c) represents the histogram of mesh
selected based on their higher pixel values and Euclidean
representation, and (d) shows its initial binary representa-
Distance (ED) between each pixel is calculated. In pattern
tion after performing threshold function. Finally, we perform
recognition and computer vision, the ED playa an important
correlation coefficient (CC) to measure the strength between
role to measure the distance between objects in the image.
pixels, which are computed after threshold function. Through
Therefore, we calculate the ED between selected image pixels
CC, select the strong correlated pixels as follows: As we know
and select the minimum distance pixels. Let 1 be the pixel
that strong correlation between two variable is (Strongr = 1)
values of cluster C1 , where C1 denotes the high-intensity
and weak correlation is defined by Weakr = 0, then e r is
value pixels. Let xi ∈ 1, where i = x1 , x2 , . . . xn number
computed as follows:
of pixels. Then ED between xi pixel is calculated as follows: P P P
n Txy − ( Tx )( Ty )
q r=p P (20)
ED(xi ) = (x2 − x1 )2 + (x3 − x2 )2 , . . . (xn − xn−1 )2
e q P
n( Tx2 ) − ( Tx )2 n( Ty2 ) − ( Ty )2
P P

X n q
ED(xi ) = (xi − xi−1 )2 (18) where, e
r denotes the correlation value of threshold image,
i=0 which is utilized as a threshold value to obtain final

VOLUME 7, 2019 46267

 M. A. Khan et al.: Optimized Method for Segmentation and Classification of Apple Diseases
FIGURE 7. SCP based segmented effects. a) Threshold image; b) strong
correlated pixel based segmented image; c) morphological operation
(remove extra regions) image.
segmented image.
( V. CLASSIFICATION
1 if T (xi ) ≥ e
r In machine learning, image classification is a difficult process
Segmg (x, y) = (21)
0 Otherwise
where, Segmg (x, y) denotes the proposed SCP based seg-
mented image. Moreover, some morphological operations
such as border removal and area removal are performed on
SCP image to remove some extra regions as shown in Fig-
ure 7. In Figure 7, the thresholded image is given in (a),
then SCP based segmented image is shown in (b), which is
improved in (c) by using morphological operations.
Finally, EM algorithm is implemented on S channel (which
is computed from Equation 15 and optimize proposed SCP
segmented image based on entropy based mutual information
(EMI). The EMI method is used for fusion of SCP segmented
image and EM segmented image. Let EM (x, y) denotes the
EM segmented image, Segmg (x, y) is SCP segmented image,
CM (x, y) ∈ egmg (x, y), EM (x, y) which is addition of both
segmented images and Sg (x, y) denotes the final optimized
image, where (x,y) denotes the number of optimized pixels.
Then the EMI method is defined as follows:
CM (x, y) = ζ (x) + ζ (y) − ζ (x, y) (22)
where, ζ (x) and ζ (y) denotes the marginal entropies, and
ζ (x, y) denotes the joint entropy of x, y, which are computed
as:
X features are extracted including color, color histogram (CH),
ζ (x) = − P(x)log2 P(x)
X
x features for improving the classification accuracy.
ζ (y) = − P(y)log2 P(y)
y 1) LBP FEATURES
XX
ζ (x, y) = − P(x, y)log2 P(x, y) (23)
x y
By putting the values of ζ (x), ζ (y), and ζ (x, y) in Equation 22
and obtain an optimized segmented image, which contains
more information as compared to original SCP and EM
image.
ζ (x, y) = ζ (x) + ζ (y) − CM (x, y)
Sg (x, y) = ζ (x) + ζ (y) − CM (x, y) (24)
where, Sg (x, y) ∈ ζ (x, y), which denotes the final optimized
segmented image. The effects of EMI method are given
in Figure 8. Moreover, some other sample results of optimized
segmentation method are shown in Figure 9. The detailed
46268
FIGURE 8. Final optimized segmentation effects. a) SCP based segmented
image; b) EM segmented image; c) optimized EMI based segmented
image; d) mapped EMI on original image.
analysis of segmentation accuracy with their corresponding
ground truth image is given in Section VI.
that may be induced by several factors such as prominent
feature extraction and selection. Four types of apple disease
classes are classified including apple Blackrot, Rust, Scab
and healthy. The classification process consists of two series
of sub-steps including feature extraction and selection. In fea-
ture extraction step, three types of features are fused and
selected the best features by employing the genetic algo-
rithm (GA) as given in Algorithm 1. The detailed description
of each step is given below.
A. FEATURE EXTRACTION
The development of pattern recognition and computer vision
applications, the texture and color features plays a major role
in classification. The LBP features also known as texture
features has been favorably employed for image classifi-
cation under several conditions such as rotation invariant
and multi-resolution constraints [23]. Moreover, the color
features are also employed in many applications such as
medicine and agriculture. However, in this article, the color
features are used for classification of apple diseases. The
color features work better with LBP features because every
disease spot has their own characteristics such as shape, color,
and texture [24]. Therefore, in this article, three types of
and LBP. The CH features are fused with color and LBP
The LBP features are popular in pattern recognition to take
the texture analysis of an image. The LBP extract the texture
information of an image by estimating their neighborhood
pixels. Let Sg (x, y) be an optimized segmented image having
size M × N , where (x, y) denotes the position of pixels in
image. Given a central pixel pc and its neighbouring pixel pg ,
then the LBP is computed as follows [23]:
g−1
X
ψ(SF)g,r = s(pg − pc )2g (25)
g=0
(
1, x ≥ 0
where, s(x) = , g and r denotes the neighbouring
0, x < 0
pixel and radius of neighbourhood, respectively. The size of
VOLUME 7, 2019
M. A. Khan et al.: Optimized Method for Segmentation and Classification of Apple Diseases

FIGURE 9. Image segmentation effects. a) HSI image; b) selected S Channel; c) SCP segmented image; d) EM
segmentation; e) optimized EMI segmentation, and f)mapped EMI on segmented spots.

extracted LBP vector is 1×59, which is later fused with color, presented red, green, and blue channel, respectively, where
and CH features as shown in Figure 10. (u, v) denotes the pixel of red channel. Then, mean, variance,
SD, skewness, kurtosis, entropy, energy, and mode of R(u, v)
2) COLOR HISTOGRAM FEATURES is calculated as follows [25]:
As G33(x, y) be an RGB enhanced image having size J × L, X X R(u, v)
where J = 256, and L = 256. Then, we perform conver- µc = (26)
u v
Z
sion of three color spaces including RGB, HSV, and HSI
(R(u, v))2
P
for CH feature extraction. Initially, we divide each image σc2 = − µ2c (27)
into their respective three channels. For example, the RGB s Z
image consists of three channels such as red, green, and P
(R(u, v))2
blue. Thereafter, we perform histogram equalization on each σc = − µ2c (28)
Z
color channel and extract 8 different feature including mean,
Z −1
variance, SD, skewness, kurtosis, entropy, energy, and mode 1 X
SKc = (R(z1 ) − µc )3 Probg (29)
for all channels. The mean of the channel describes its general σc3 z
1
brightness, however, the variance and SD are used to find µc
out how each pixel in the image varies from the neighboring Modec = (30)
σc × SKc
pixels or central pixels, and contrast of an image, respectively.
Z −1
Moreover, the skewness finds the asymmetry in their intensity X
Entropyc = − Prob(R(u, v)). log2 Prob(R(u, v)) (31)
values, entropy explains that how many bits are needed to
z1 =0
code the image data, and energy tells about the smoothing
Z −1
pixels in the given channel. These selected features are cal- X
Energyc = [Prob(R(u, v))]2 (32)
culated as follows:
z1
Let G33(x, y) is a three-dimensional enhanced RGB
G33(x) G33(y) Z −1 (R(u,v))i −µc )
image, then R(u, v) = G33(x,y,z) , G(u, v) = G33(x,y,z) , X
Z
Kurc = (33)
G33(z)
and B(u, v) = G33(x,y,z) , where R(u, v), G(u, v), and B(u, v) z1
σ 4
c

VOLUME 7, 2019 46269

 M. A. Khan et al.: Optimized Method for Segmentation and Classification of Apple Diseases
FIGURE 10. A detailed flow diagram of feature fusion and selection.
where, µc is mean feature for red channel, Z denotes the total
number of pixels in R(u, v) channel, σc2 is variance feature
for red channel, σc2 is SD feature for red channel, SKc is
skewness feature, z1 ∈ (u, v), Probg is probability of gray
levels Z −1, Modec is mode feature for red channel, Entropyc
is entropy feature, Prob(R(u, v)) probability of red channel
pixels, Energyc is energy feature for red channel, and Kurc
is kurtosis feature for red channel. Hence, for one channel,
total 8 features are extracted. Similarly, the length of all three
color spaces, the feature vector (FV) is 1 × 72 as shown
in Figure 10. Finally, the simple color features are extracted
from the original RGB image. To follow above CH features
extraction process, 8 similar feature parameters are extracted
for each channel. The size of extracted feature vector for one
channel is 1 × 8 and length of final color FV is 1 × 72, which
is later fused with CH and LBP feature vector.
For feature fusion, a high-value feature based parallel
fusion is employed. The fusion process is completed in two
steps: a) same length of each feature vector, and b) high-value
feature based fusion. In the first step, make the same size of
each feature vector based on ‘‘0’’ appending. After, that each
index of given feature vectors is compared to each other and
the higher-value feature is selected for a fused vector. The
length of a final fused vector is 1 × 72. The detailed flow of
color feature extraction and fusion is given in Figure 10.
 tion is defined as follows:
ψ(CH )sl if ψ(CH )i ≥ ψ(CL)i & ψ(SF)i

FV (i) = ψ(CL)sl if ψ(CL)i ≥ ψ(CH )i & ψ(SF)i
ψ(SF)sl if ψ(SF)i ≥ ψ(CH )i &ψ(CL)i


(34)
where, FV (i) denotes the final fused feature vector, ψ(CH )sl ,
ψ(CL)sl , and ψ(SF)sl denotes the selected color histogram,
color, and LBP feature, respectively.
46270
B. FEATURE SELECTION
For computational purposes, images are represented by col-
lection of features. To build an efficient classification model,
it is essential that features contain much information and
number of features are kept as small as possible. However,
it is difficult to distinguish that which features are appropriate
for accurate classification of images. Because, the extracted
features are large in number and it contains many irrelevant
features, which not only degrade the system accuracy but
also effect on the system execution time. To remove the
adverse consequences of unnecessary and redundant features,
many feature selection techniques have been proposed such
as competitive swarm optimizer based algorithm [26].
As given FV (i) is fused feature vector having dimension
1 × 72. Let ψ(OP(i)) denotes the optimized feature vec-
tor having dimension A × B, where A denotes the index
of extracted feature and B denotes the length of optimized
feature vector. Generally, the GA is comprised of five major
steps including initialization, crossover, mutation, selection,
and fitness function. In initialization step, the selected num-
ber of iterations (It = 500), number of populations (NP =
20), crossover rate (CP = 0.4), mutation rate (MP = 0.001),
and selection of parent pressure is β = 6. Amongst, sev-
eral crossover techniques, we select the uniform crossover
technique [27] with crossover rate 0.4. The crossover func-
ψCR = CROSS(h1 , h2 )
h1 = α × f1 + (1 − α) × f2
h2 = α × f2 + (1 − α) × f1 (35)
where, h1 , h2 denotes the given input features, and f1 , f2
are new offspring individuals after crossover operation.
In the next step, the offsprings are mutated using uniform
VOLUME 7, 2019
M. A. Khan et al.: Optimized Method for Segmentation and Classification of Apple Diseases

Algorithm 1 Features Fusion and Selection TABLE 2. Brief description of selected training and testing images of
1: Input : ψ(CL), ψ(CH ), ψ(SF)
various apple leaf diseases.

2: Output : ψ(Op) F where ψ(Op) is optimized final

selected feature vector
3: It = 500 F where It is number of iterations
4: NP = 20 F where NP is number of Populations
5: CP = 0.4 F where CP is crossover rate
6: MP = 0.001 F where MP is mutation rate
7: β = 6 S
where, i ∈ 72, pr(i) = exp(−β × Wpc ), Sp denotes the sorted
8: ψ(CL) ← CL1 , CL2 , CL3 , . . . CLn F where, ψ(CL) population, and Wc denotes the last value of population.
denotes the color feature vector Finally, the fitness function is performed to select the best
9: ψ(CH ) ← CH1 , CH2 , CH3 , . . . CHm F where, ψ(CH )
features for classification. The fitness function is defined as
denotes the color histogram feature vector follows:
10: ψ(SF) ← SF1 , SF2 , SF3 , . . . SFo F where, ψ(SF)
denotes the LBP feature vector Fitness(ψ(OP(i))) = ACC(FV ) − min ψ(λ1 ) (37)
λ1
11: Length(ψ) ← Length of max feature vector
12: Length(ψ) ← max(ψ(CL), ψ(CH ), ψ(SF)) where, λ1 ∈ λ which denotes the selected feature by employ-
13: ψ(CL)(length(ψ(CL)) + 1 : Length(ψ)) = 0 F Its ing Roulette Wheel, and ψ(OP(i)) denotes the final opti-
make the length of features equal mized feature vector, which is later classified by One-Vs-All
14: ψ(CH )(length(ψ(CH )) + 1 : Length(ψ)) = 0 multi-class support vector machine (M-SVM) [30].
15: ψ(SF)(length(ψ(SF)) + 1 : Length(ψ)) = 0
16: For i = 1 to Length(ψ) VI. EXPERIMENTAL RESULTS AND DISCUSSION
17: Fv(i) ← max(ψ(CL), ψ(CH ), ψ(SF)) F Save max The proposed method is evaluated on Plant Village dataset,
feature from each iteration which contains thousands of images of several types of
18: i ← i + 1 classes including citrus, apple, grapes, corn, tomato, and
19: ENDFOR few more. In this article, we select four apple leaf disease
20: FV (i) ← Fused Feature vector classes including apple scab, apple rust, apple black rot,
21: Pi ← Initialize Population and healthy images as shown in Figure 11 and brief detail
22: Evaluate Pi is presented in Table. 2 including total number of images
23: For j = 1 to It in each class, number of training and testing samples, and
24: For k = 1 to length(FV ) capturing condition like controlled and uncontrolled environ-
25: Spop ← Select parrent for population ment. The background is same which is used at the time of
26: Xoff ← Crossover(Xoff (CP)) F perform crossover image capturing and also all images includes pepper noise
27: Xmut ← Mutate(Xoff , MP) F perform mutation which generates irrelevant features and make the classifi-
28: Evaluate Offspring cation process difficult. The validation of proposed system
29: Popt+1 ← actual population through replacement is calculated through eight classification methods including
30: Pop&Xoff0 complex tree (CT), linear-SVM (L-SVM), quadratic-SVM
31: Fitness ← accuracy(FV ) − minψ(λ1 ) F perform (Q-SVM), fine-KNN (F-KNN), weighted-KNN (W-KNN),
λ1
fitness function ensemble boosted tree (EBT), ensemble discriminant anal-
32: λ1 ∈ λ ysis (ESDA), and M-SVM. The M-SVM is used as a Base
33: ENDFOR classifier and compared its performance with other selected
34: j←j+1 F iterate upto all features classifiers. The classification performance is calculated based
35: ψ(Op(i)) F Optimized feature vector on eight performance parameters named as sensitivity (Sen),
36: ENDFOR specificity (Spec), FNR, FPR, precision (Prec), Informedness
(INF), G-Measure (G-M), and accuracy (Acc). Moreover,
the system performance is also measured by their execution
time. The evaluation is analyzed by three different test such
mutation [28] with MP = 0.001. The mutation operation as without feature selection algorithm, PCA based feature
is found to some unique individuals, which are generated reduction, and proposed feature selection algorithm. All the
by crossover. Then, offspring is evaluated and the actual results are done on Matlab 2017a using Intel Core I7 with
population is selected through the replacement of individuals. 16GB of RAM and 8GB GPU.
The selection is done by Roulette Wheel [29], which is based
on their probability values. The probability is defined as A. SEGMENTATION RESULTS
follows:
The proposed symptoms segmentation method is tested
pri on total of 300 images (100 apple scab, 100 apple rust,
Prob(i) = P (36)
(pr(i)) and 100 apple rot) which are collected from a database

VOLUME 7, 2019 46271

 M. A. Khan et al.: Optimized Method for Segmentation and Classification of Apple Diseases
FIGURE 11. Annotated sample for each selected apple disease and healthy leaf.
FIGURE 12. Sample images for segmentation accuracy.
(Plant Village) and few of sample images are shown in
Figure 12. The ground truth of selected images is drawn from
an expert in this area from Agriculture University Fasilabad,
Pakistan. The accuracy of segmented images is computed by
their corresponding ground truth images and few of segmen-
tation results in numerical form are also presented in Table 3.
The average accuracy is 93.916% which is good as compare
to other methods as plotted in Figure 14. The maximum
segmentation accuracy on our method is achieved as 97.79%.
Moreover, the sample visual segmentation results with their
corresponding ground truth images are shown in Figure 13.
The above discussion clearly showed that the proposed seg-
mentation method performs well on selected images.
B. TEST 1: WITHOUT FEATURE SELECTION
In this test, the color, color histogram, and texture features
are extracted using proposed disease segmentation algo-
rithm. The extracted features are fused one vector based on
parallel method. The fused vector is fed to classifiers and
classification results are obtained. For experimental results,
46272
we opt a ratio of 50:50 ratio and results are evaluated using
10-fold cross-validation (10CV). The best achieved accuracy
of 92.9% for M-SVM as given in Table 4. The classification
accuracy of M-SVM is also verified through Table 5. More-
over, the graphical comparison of classification methods is
given in Figure 14, which shows that the M-SVM perform
significantly well for Test 1 in terms of specificity, accuracy,
INF, and G-measure. Moreover, the testing execution time
of M-SVM is 7.862 seconds, which is better as compared
to other classification methods. The worst execution time is
69.260 seconds for ESDA but its classification rate is 91.50%,
which is better as compared to CT, L-SVM, F-KNN, W-KNN,
and EBT.
C. TEST 2: PCA BASED FEATURE REDUCTION
In this test, parallel fusion based features are combined. The
features are reduced by utilizing PCA and fed to classi-
fiers to obtained the classification results. For experimental
results, we opt a ratio of 50:50 results are computed using
10CV. The maximum classification accuracy is 94.3% and
VOLUME 7, 2019
M. A. Khan et al.: Optimized Method for Segmentation and Classification of Apple Diseases
TABLE 3. Accuracy results for proposed optimized segmentation method.
TABLE 4. Classification results for without feature selection algorithm.
TABLE 5. Confusion matrix of without feature selection algorithm.
the error rate is 5.7%, which is achieved on M-SVM as given
in Table 6. Moreover, some other parameters for M-SVM are
including INF, specificity, sensitivity, G-M is 0.849, 92.49%,
92.50%, and 93.37%, respectively. The classification accu-
racy of M-SVM is confirmed by their confusion matrix given
in Table 7. Moreover, the graphical comparison of classifi-
cation methods is given in Figure 15, which shows that the
M-SVM perform significantly well for Test 2 in terms of
specificity, accuracy, INF, and G-measure. Moreover, the test-
ing execution time of M-SVM is 3.139 seconds, which is
better as compared to other classification methods. The worst
execution time is 20.886 seconds for EBT with classification
rate is 89.50%. From above discussion and results, it clearly
shows that the PCA based feature reduction results are good
as compare to Test 1 (feature fusion and without feature
selection). In test 1, the maximum classification accuracy of
M-SVM is 92.90% with execution time 7.862 seconds. How-
ever, the Test 2 maximum classification accuracy is 94.3%,
with execution time 3.139. It shows that Test 2 is performing
well as compare to Test 1 and also reduce the execution time.
D. TEST 3: PROPOSED METHOD RESULTS
The proposed method results are presented in this section.
The fused vector is fed to GA for selection of robust and most
VOLUME 7, 2019
TABLE 6. Classification results for PCA based feature reduction.
TABLE 7. Confusion matrix for PCA based feature reduction.
TABLE 8. Classification results of proposed method.
TABLE 9. Confusion matrix for proposed method.
prominent features. For experimental results, we opt a ratio
of 50:50 and results are analyzed using 10CV. The maximum
classification accuracy of the proposed method is 97.20%
and the error rate is 2.8% as given in Table 8. Moreover, the
M-SVM performance is calculated by some other parameters
such as INF (94.2), specificity (96.90), sensitivity (97.15),
G-M (96.94), and FPR (0.01), respectively. The best accuracy
is also verified through Table. 9 in the form of confusion
matrix. Moreover, the graphical comparison of classification
methods is given in Figure 16, which shows that the M-SVM
perform significantly well for Test 3 (proposed method) in
terms of specificity, accuracy, INF, and G-measure. More-
over, the test execution time of M-SVM is 0.651 seconds,
which is better as compared to other classification methods.
The worst execution time is 5.659 seconds for L-SVM with
classification result of 93.6%. From above discussion and
results, it clearly shows that the proposed method outper-
forms. As compared to Table. 4 and Table. 6, it’s clearly
shown that the proposed method (Test 3) results outperform
the others. Moreover, the execution time of proposed method
compared to Test 1 and Test 2 is shown in Figure 17.
46273
 M. A. Khan et al.: Optimized Method for Segmentation and Classification of Apple Diseases

FIGURE 13. Comparison of proposed segmented effects with ground truth images. a)
Original RGB image; b) selected S channel; c) optimized EMI segmentation; d) mapped
on original image, and e) ground truth image.

FIGURE 14. Graphical comparison of M-SVM with other classification methods in terms of accuracy,
specificity, Informedness, and G-Measure for Test 1.

E. DISCUSSION in Figure 8 and 9. In classification step, three types of fea-

Three pipeline procedure including image enhancement,
image segmentation, and classification are performed in this
work as shown in Figure 1. In the first step, the preprocessing
is performed on query image to improve the contrast of
disease part as shown in Figure 2, 3, and 4. In the second
step, an SCP based segmentation algorithm is implemented
as shown in Figure 5, 6, and 7. The proposed SCP image is
optimized by EM segmentation and their effects are given
tures are fused including color, CH, and LBP and select
the best features based on GA as flow diagram is given
in Figure 10. For qualitative analysis, four apple classes are
collected including apple scab, apple black rot, apple rust,
and healthy images. Before classification results, the seg-
mentation accuracy is obtained based on their corresponding
ground truth image. For this purpose, 30 images are randomly
selected and their results are given in Table 2, and Figure 11.

46274 VOLUME 7, 2019

M. A. Khan et al.: Optimized Method for Segmentation and Classification of Apple Diseases

FIGURE 15. Graphical comparison of M-SVM with other classification methods in terms of accuracy,
specificity, Informedness, and G-Measure for Test 2.

FIGURE 16. Graphical comparison of M-SVM with other classification methods in terms of accuracy,
specificity, Informedness, and G-Measure for Test 3.

For classification results, three different experiments are done
on select dataset such as without feature selection, PCA based
feature reduction, and proposed feature selection based anal-
ysis. The maximum classification accuracy of Test 1 is 92.9%
as presented in Table 3 and Table 4. In Test 2, the PCA based
feature reduction method performs well as compare to Test 1
and achieved maximum classification accuracy is 94.3% as
given in Table 5 and Table 6. The proposed achieved maxi-
mum classification accuracy 97.20% as presented in Table 7,
which is confirmed by their confusion matrix in Table 8.
The proposed method performs significantly well as com-
pare to Test 1 and Test 2 in terms of execution time, INF,
G-M, Sensitivity, and Precision. Moreover, the graphical rep-
resentation of all three Tests is shown in Figure 12, 13, and 14.
Existing well-known techniques such as Yolo, RCNN,
F-RCNN, and SSD models show improved performance in
object detection for both real-time and static image. For the
case when the object/s of interest is too small, the overall per-
formance will be compromised in terms of overall accuracy.
In [34], authors validated YOLO CNN model on the traffic
signs and concluded that a set of stop signs which are quite far,
as well as small in size are not easy to detect, hence, showed
only 30% accuracy. Whereas, in this work, our primary objec-
tive is to detect the region of interest (infected) and then
extract its features for recognition. Due to a fact, the features
from cropped images give better accuracy, as the irrelevant
information (background) is discarded, and only the rele-
vant information is considered. For the aforementioned CNN

VOLUME 7, 2019 46275

 M. A. Khan et al.: Optimized Method for Segmentation and Classification of Apple Diseases
FIGURE 17. Execution time of all three experiments for M-SVM.
TABLE 10. Comparison with existing methods for apple disease
classification.
models, a large number of ground truth images are required
for training, which in our used dataset, are not publicly avail-
able. Moreover, our algorithm detects minuscule symptoms
from the leaf images more accurately - compared to set of
CNN models, as well as other algorithms like thresholding
etc. The recognition accuracy of existing techniques are also
presented in Table 10. For comparison, we are only consid-
ering the methods tested on plant village dataset. Recently,
Khan et al. [6] presented an automated approach for fruit
diseases by achieving an accuracy of 98.10% on apple black
rot and for apple scab, 96.90%. In this work, we achieve an
average accuracy of 98.10% on of apple black rot, on apple
scab its 97.30%, and 94.6% on apple rust. From the results,
it can comfortable concluded that the proposed method out-
performs existing methods - validated on same dataset.
VII. CONCLUSION
In this article, an optimized automated computer-based
method is proposed and validated for recognition of apple
diseases. The lesion spot contrast stretching, lesion segmen-
tation, and prominent features selection & recognition steps
are utilized. The contrast of infected spot is enhanced and
46276
segmentation is performed by the proposed SCP method.
The performance of the proposed SCP method is further
optimized by EMI approach. Then, multiple features are
extracted and fused by using a parallel method. A Genetic
algorithm is applied to select the best features, which are later
utilized by M-SVM for classification. The M-SVM achieved
a classification accuracy of 92.9%, 94.30%, and 97.20% for
without feature selection (Test 1), PCA based feature reduc-
tion (Test 2), and proposed features selection method (Test 3),
respectively. The results show that the selection method gives
better performance in terms of accuracy and execution time.
The deep learning gives promising results and has a large
potential to further improves. Recently, deep learning has
been successfully applied in various computer vision field
and also registered in agriculture for a variety of plants and
fruits classification. Through deep learning, data is repre-
sented through a hierarchy of convolutions [32], [33], [35].
In the future, deep learning based features are extracted
through cropped symptoms regions. Through deep learning,
there is no need to perform preprocessing, therefore this step
can be skipped which further improves the computational
cost.
ACKNOWLEDGEMENT
The authors extend their appreciation to the Deanship of
Scientific Research at King Saud University for funding this
work through research group NO (RG-1438-034).
REFERENCES
[1] S. Savary, A. Ficke, J-N. Aubertot, and C. Hollier, ‘‘Crop losses due to
diseases and their implications for global food production losses and food
security,’’ Food Secur., vol. 4, no. 4, pp. 519–537, 2012.
[2] K. R. Gavhale, and U. Gawande, ‘‘An overview of the research on plant
leaves disease detection using image processing techniques,’’ J. Comput.
Eng., vol. 16, no. 1, pp. 10–16, 2014.
[3] B. J. Samajpati, and S. D. Degadwala, ‘‘Hybrid approach for apple
fruit diseases detection and classification using random forest classi-
fier,’’ in Proc. Int. Conf. Commun. Signal Process. (ICCSP), Apr. 2016,
pp. 1015–1019.
[4] Z. Chuanlei, Z. Shanwen, Y. Jucheng, S. Yancui, and C. Jia, ‘‘Apple leaf
disease identification using genetic algorithm and correlation based feature
selection method,’’ Int. J. Agricult. Biol. Eng., vol. 10, no. 2, pp. 74–83,
2017.
[5] A. Camargoa and J. S. Smith, ‘‘An image-processing based algorithm
to automatically identify plant disease visual symptoms,’’ Biosyst. Eng.,
vol. 102, no. 1, pp. 9–21, 2009.
[6] M. A. Khan et al., ‘‘CCDF: Automatic system for segmentation and
recognition of fruit crops diseases based on correlation coefficient and
deep CNN features,’’ Comput. Electron. Agricult., vol. 155, pp. 220–236,
Dec. 2018.
[7] L. J. Rozario, T. Rahman, and M. S. Uddin, ‘‘Segmentation of the region
of defects in fruits and vegetables,’’ Int. J. Comput. Sci. Inf. Secur., vol. 14,
p. 399, May 2016.
[8] M. Sharif, M. A. Khan, Z. Iqbal, M. F. Azam, M. I. U. Lali, and
M. Y. Javed, ‘‘Detection and classification of citrus diseases in agriculture
based on optimized weighted segmentation and feature selection,’’ Com-
put. Electron. Agricult., vol. 150, pp. 220–234, Jul. 2018.
[9] A. T. Sapkal and U. V. Kulkarni, ‘‘Comparative study of leaf disease
diagnosis system using texture features and deep learning features,’’ Int.
J. Appl. Eng. Res., vol. 13, no. 19, pp. 14334–14340, 2018.
[10] A. M. AlShahrani, M. A. Al-Abadi, A. S. Al-Malki, A. S. Ashour, and
N. Dey, ‘‘Automated system for crops recognition and classification,’’
in Computer Vision: Concepts, Methodologies, Tools, and Applications.
Pennsylvania, PA, USA: IGI Global, 2018, pp. 1208–1223.
VOLUME 7, 2019
M. A. Khan et al.: Optimized Method for Segmentation and Classification of Apple Diseases
[11] P. Moghadam, D. Ward, E. Goan, S. Jayawardena, P. Sikka, and
E. Hernandez, ‘‘Plant disease detection using hyperspectral imaging,’’ in
Proc. Int. Conf. Digit. Image Comput. Techn. Appl. (DICTA), Dec. 2017,
pp. 1–8.
[12] K. P. Ferentinos, ‘‘Deep learning models for plant disease detection and
diagnosis,’’ Comput. Electron. Agricult., vol. 145, pp. 311–318, Feb. 2018.
[13] S. R. Dubey and A. S. Jalal, ‘‘Apple disease classification using color,
texture and shape features from images,’’ Signal, Image Video Process.,
vol. 10, no. 5, pp. 819–826, 2016.
[14] S. R. Dubey and A. S. Jalal, ‘‘Detection and classification of apple fruit
diseases using complete local binary patterns,’’ in Proc. 3rd Int. Conf.
Comput. Commun. Technol., Nov. 2012, pp. 346–351.
[15] E. Omrani, B. Khoshnevisan, S. Shamshirband, H. Saboohi, N. B. Anuar,
and M. H. N. M. Nasir, ‘‘Potential of radial basis function-based support
vector regression for apple disease detection,’’ Measurement, vol. 55,
pp. 512–519, Sep. 2014.
[16] M. Shuaibu, W. S. Lee, Y. K. Hong, and S. Kim, ‘‘Detection of apple
marssonina blotch disease using particle swarm optimization,’’ Trans.
Amer. Soc. Agricult. Biol. Eng., vol. 60, no. 2, pp. 303–312, 2017.
[17] V. Singh and A. K. Misra, ‘‘Detection of plant leaf diseases using image
segmentation and soft computing techniques,’’ Inf. Process. Agricult.,
vol. 4, no. 1, pp. 41–49, 2017.
[18] J. G. A. Barbedo, ‘‘A new automatic method for disease symptom segmen-
tation in digital photographs of plant leaves,’’ Eur. J. Plant Pathol., vol. 147,
no. 2, pp. 349–364, 2017.
[19] H. Ali, M. I. Lali, M. Z. Nawaz, M. Sharif, and B. A. Saleem, ‘‘Symptom
based automated detection of citrus diseases using color histogram and
textural descriptors,’’ Comput. Electron. Agricult., vol. 138, pp. 92–104,
Jun. 2017.
[20] V. Ugarte, U. Sinha, V. Malis, R. Csapo, and S. Sinha, ‘‘3D multimodal
spatial fuzzy segmentation of intramuscular connective and adipose tissue
from ultrashort TE MR images of calf muscle,’’ Magn. Reson. Med.,
vol. 77, no. 2, pp. 870–883, 2017.
[21] C. R. Gonzalez, ‘‘Richard Eugene woods,’’ Digital Image Processing.
vol. 85, 2007.
[22] Y. Mitani and Y. Hamamoto, ‘‘A consideration of pan-sharpen images by
HSI transformation approach,’’ in Proc. Annu. Conf. SICE, Aug. 2010,
pp. 1283–1284.
[23] O. Timo, M. Pietikäinen, and T. Mäenpää, ‘‘Multiresolution gray-scale and
rotation invariant texture classification with local binary patterns,’’ IEEE
Trans. Pattern Anal. Mach. Intell., vol. 24, no. 7, pp. 971–987, Jul. 2002.
VOLUME 7, 2019
[24] S. Zhanga, X. Wu, Z. Youa, and L. Zhang, ‘‘Leaf image based cucumber
disease recognition using sparse representation classification,’’ Comput.
Electron. Agricult., vol. 134, pp. 135–141, Mar. 2017.
[25] S. Sergyan ‘‘Color histogram features based image classification in
content-based image retrieval systems,’’ in Proc. 6th Int. Symp. Appl.
Mach. Intell. Inform., Jan. 2008, pp. 221–224.
[26] S. kai, G. Cheng, and Y. Jin ‘‘Feature selection for high-dimensional
classification using a competitive swarm optimizer,’’ Soft Comput., vol. 22,
no. 3, pp. 811–822, 2018.
[27] S. Kumara, D. E. Goldberg, and G. Kendall. ‘‘Genetic algorithms,’’ in
Search Methodologies. New York, Ny, USA: Springer, 2014, pp. 93–117.
[28] S. N. Sivanandam, and S. N. Deepa. Introduction to Genetic Algorithms.
New York, NY, USA: Springer, 2007.
[29] A. Lipowski and D. Lipowska, ‘‘Roulette-wheel selection via stochastic
acceptance,’’ Phys. A, Stat. Mech. Appl., vol. 391, no. 6, pp. 2193–2196,
2012.
[30] Y. Liu and Y. F. Zheng, ‘‘One-against-all multi-class SVM classifica-
tion using reliability measures,’’ in Proc. IEEE Int. Conf. Neural Netw.,
Jul. 2005, pp. 849–854.
[31] G. Wang, Y. Sun, and J. Wang, ‘‘Automatic image-based plant disease
severity estimation using deep learning,’’ Comput. Intell. Neurosci., vol. 9.
no. 4, pp. 2193–2196, 2017.
[32] A. Kamilaris and F. X. Boldú, ‘‘Deep learning in agriculture: A survey,’’
Comput. Electron. Agricult., vol. 147, pp. 70–90, Apr. 2018.
[33] T. Wiesner-Hanks et al., ‘‘Image set for deep learning: Field images of
maize annotated with disease symptoms,’’ BMC Res. Notes, vol. 11, no. 1,
p. 440, 2018.
[34] J. Lu, H. Sibai, E. Fabry, and D. Forsyth. (2017). ‘‘No need to worry
about adversarial examples in object detection in autonomous vehicles.’’
[Online]. Available: https://arxiv.org/abs/1707.03501
[35] Z. Iqbal, M. A. Khan, M. Sharif, J. H. Shah, M. H. Rehman, and K. Javed,
‘‘An automated detection and classification of citrus plant diseases using
image processing techniques: A review,’’ Comput. Electron. Agricult.,
vol. 153, pp. 12–32, Oct. 2018.
Authors’ photographs and biographies not available at the time of
publication.
46277