Traitement du Signal

Vol. 37, No. 6, December, 2020, pp. 1093-1101

Journal homepage: http://iieta.org/journals/ts

A Novel Convolutional Neural Network Based Model for Recognition and Classification of
Apple Leaf Diseases
Divakar Yadav*, Akanksha, Arun Kumar Yadav

Department of Computer Science and Engineering, National Institute of Technology, Hamirpur (HP) 177005, India

Corresponding Author Email: [email protected]

https://doi.org/10.18280/ts.370622 ABSTRACT

Received: 3 October 2020 Plants have a great role to play in biodiversity sustenance. These natural products not only
Accepted: 11 December 2020 push their demand for agricultural productivity, but also for the manufacturing of medical
products, cosmetics and many more. Apple is one of the fruits that is known for its excellent
Keywords: nutritional properties and is therefore recommended for daily intake. However, due to
plants, apple, contrast stretching, fuzzy c- various diseases in apple plants, farmers have to suffer from a huge loss. This not only causes
means, CNN, disease diagnosis severe effects on fruit’s health, but also decreases its overall productivity, quantity, and
quality. A novel convolutional neural network (CNN) based model for recognition and
classification of apple leaf diseases is proposed in this paper. The proposed model applies
contrast stretching based pre-processing technique and fuzzy c-means (FCM) clustering
algorithm for the identification of plant diseases. These techniques help to improve the
accuracy of CNN model even with lesser size of dataset. 400 image samples (200 healthy,
200 diseased) of apple leaves have been used to train and validate the performance of the
proposed model. The proposed model achieved an accuracy of 98%. To achieve this
accuracy, it uses lesser data-set size as compared to other existing models, without
compromising with the performance, which become possible due to use of contrast
stretching pre-processing combined with FCM clustering algorithm.

1. INTRODUCTION While researchers have worked rigorously to identify plant

diseases using different methods such as RNA/DNA, sensor
The Indian economy is heavily dependent on efficient techniques, etc. [5] but the field of machine vision to identify
agriculture. The detection of diseases in plants therefore plays manifestations of fruit leaf diseases is still less examined.
an important role in agriculture [1]. The use of automated Apples are one of the widely consumed fruits, a great source
disease detection techniques is advantageous for the fast of phytochemicals mostly expressing pertinent antioxidant
identification of diseases in plants [2]. For instance, black rot abilities in vitro, and scientific studies have related apple
is one of the most prevalent and serious diseases that plagues ingestion to a lower chance of certain cancers, cardiovascular
apple trees. They appear as brown spots, which expand in disease, asthma, and diabetes [6, 7]. The consolidated list of
concentrated circles and finally turn black, decaying the fruits. abbreviations used in the manuscript is as shown in Table 1.
Later, the disease spreads to the roots of the tree causing
cancers that can ultimately kill the tree. Early-stage detection Table 1. Acronyms used in the manuscript
of these diseases in such situations could have been helpful.
To prevent large losses, different techniques for diagnosing Symbols Abbreviations
diseases have been developed in the past. Techniques CNN convolutional neural network
developed in microbiology and immunology offer correct FCM Fuzzy C-means
recognition of the causative agents. Nonetheless, for many SVM Support Vector Machine
farmers, these approaches are inaccessible and require a PSO Particle Swarm Optimization
CFS correlation-based feature selection
thorough knowledge of the region or a large amount of money CLAHE Contrast Limited Adaptive Histogram Equalization
and energy to carry out. As per the United Nations Food and GLCM Gray Level Co-Occurrence Matrix
Agriculture Organization, most farms in the world are small PHOG Pyramid Histogram of Oriented Gradients
and managed by families in developing countries such as India GA genetic algorithm
[3]. Such families grow food for a large proportion of the
population of the country. Even so, hunger and food scarcity The main contributions in this work are stated as follows:
are not unusual and, market access and resources are 1. A novel convolutional neural network based model for
constrained. For the above reasons, much work has been done recognition and classification of apple leave disease is
in order to develop methods that are sufficiently reliable and proposed. The proposed model utilizes contrast stretching
available to the majority of farmers. The techniques of digital based pre-processing and fuzzy c-means clustering for image
image processing increase the chance of early identification of segmentation. Both these approaches boost the performance of
diseases in plants, so that the required preventive steps can be CNN classier even on lesser size of training data as compared
taken [4]. to other state of the art methods.

1093
 2. A comprehensive discussion on the existing work is implemented by SVM classifier. Ashourloo et al. [19] utilized
presented to elaborate the research gaps. regression techniques for the identification of wheat plant rust
3. Extensive computer simulations are performed to disease. Later, they also evaluated the effect of data set on
determine the effectiveness of the proposed system. A results. Ali et al. [20] used the ∆E color difference algorithm
benchmark dataset (Kaggle) which is composed of 4-types of to isolate the affected area from the leaf as well as color
apple leaves are used for simulations. Simulation result reveals histogram and compositional features to identify diseases.
that the proposed system showed competitive performance They applied principal components analysis for the features
over the other state-of-the-art methods. set dimension reduction and Bagged tree classifier for
classification. Kaur et al. [21] used desegregated Particle
Swarm Optimization (PSO) technique and support vector
2. RELATED WORK machine (SVM) for the identification and classification of
plant leaf diseases. The prime motive of this study was to
Traditional ways for identifying as well as analyzing the identify portion of the leaf, affected by the disease and a stable
diseases in fruit leaves are manual. However, these manual portion of the leaf. Ma et al. [22] employed comprehensive
processes take time, are cumbersome and also very subjective color function and its method of detection that can segment
[8]. Several methods have been developed in recent years images of disease spots recorded under real field
incorporating computer vision to detect and identify circumstances. The approach ensures reliable feedback into
agricultural and horticultural crop diseases to address the detection of CNN-based disease recognition.
manual techniques issues [9, 10]. Image collection, retrieval of Mondal et al. [23] used forty-three morphological
features, filtering of features and classification analysis with characteristics of okra and bitter gourd leaves to identify
parametric or non-parametric statistics are fundamental steps symptoms of disease from their images. In this approach,
in those processes. Image processing techniques and feature set collection was rendered utilizing Pearson
classification mechanisms are the main concern for the correlation coefficient and the entropy-based discretization
efficient functioning of the computer vision system. was used to improve the classification success rate. Zhang et
Research on the identification of plant disease using al. [24] utilized various useful features that were chosen by
machine learning is on the rise. The main reason it may be that combining the merits of the genetic algorithm (GA) with
expert eye observation of scientists has often proved to be very correlation-based feature selection (CFS). Here, GA and CFS
impractical for such systems and moreover, constant played a key role in decreasing the dissensions of the feature
surveillance is needed, which is very costly when dealing with space. Lastly, SVM classifier was used for diseases
large farms [11]. In some areas, farmers do not have identification. Singh et al. [25] described a method of image
appropriate facilities or even the knowledge that they can seek segmentation using GA and later classification was done by
experts. Under these conditions, automatic disease detection using SVM. B. Liu et al. [26] proposed a model for accurate
by seeing the signs on the leaves of plants makes the system identification of apple leaf disorder that requires producing
much faster, easier and cheaper. This also promotes machine many pathological representations. Finally, they developed a
vision to offer automated image based process control, novel architecture of an AlexNet based deep CNN for disease
examination and robot assistance [12]. detection. A computational network of CNNs was proposed by
Selvaraj et al. [13] suggested a four step scheme: first an Dechant et al. [27] to tackle the limitations of restricted data
RGB image color transformation system is generated for the and the countless variations that occur in field grown leaf
input, the green pixels are then covered and replaced using images. Multiple CNNs were trained to identify small image
different threshold values preceded by segmentation. For areas, and their predictions were compiled into independent
useful segments, texture statistics are calculated and the heat maps that are then fed into a final CNN trained model to
derived features are finally passed to the SVM classifier. identify the entire image as diseased or not.
Pujari et al. [14] proposed Support Vector Machine and Hanson et al. [28] proposed a new method for the detection
Artificial Neural Network based identification and of plant diseases using a deep convolutional neural network
classification of fungal disease in cereals. The regions trained as well as fine-tuned to suit appropriately with plant
concerned are segmented using k-means segmentation. Color leaves database that was independently collected for various
texture characteristics are extracted from affected regions and plant diseases. Yao et al. [29] proposed an efficient three-layer
then used as classifier inputs. Vishnu et al. [15] used K-means discovery system for the characterization of different stages of
clustering technique for leaf segmentation and then calculated growth of white-backed planthoppers on rice crops in paddy
the texture features for the segmented infected objects. Lastly, farmlands. Sethy et al. [30] proposed fuzzy Logic together
the derived features were processed through neural network with K-means segmentation method to measure the extent of
model. Muthukannan and Latha [16] proposed a novel solution the disease in rice crops. Fuentes et al. [31] proposed a deep-
to image segmentation, called PSO. PSO is an efficient, self- learning method to identify diseases as well as pests in tomato
regulating unsupervised algorithm that is used for improved using photographs recorded in-place with various resolutions
segmentation and extraction of features. The hybrid by camera devices. The research illustrated the efficiency of
characteristic coefficients were then obtained from the co- deep meta-architectures and characteristic extractors.
occurrence gray level matrices of various leaves. Chung et al. Sunny et al. [32] proposed two-stage solution to enhance the
[17] have suggested an approach using support vector machine image clarity. The primary stage uses Contrast Limited
(SVM) classifiers to differentiate healthy and Bakanae- Adaptive Histogram Equalization (CLAHE) to pre-process the
infected rice seedlings. leaf image, followed by segmentation using K-mean clustering
Zhang et al. [18] suggested a technique for the identification and extraction of texture characteristics via statistical Gray
of cucumber disease based on decomposition of the global- Level Co-Occurrence Matrix (GLCM). The second stage
local single value to increase the detection rate. The utilizes support vector machine to identify the plant as healthy
classification of the unidentified disease leaf picture was or ill. Zhang et al. [33] suggested clustering algorithm to split

1094
the color-diseased image of the leaf into many small super- model for diseased cucumber images and achieved accuracy
pixels and then used K-means clustering method to fragment of 88.05% Waheed et al. [42] proposed DenseNet, a dense
the image of every super-pixel of the lesion. Finally, the convolutional neural network based model for recognition and
Pyramid Histogram of Oriented Gradients (PHOG) identification of diseases in corn leaves. They claimed that
functionality was derived from three color attributes of each their proposed method uses significantly lesser parameters as
fragmented lesion image as well as its grayscale image, and a compared to other state-of-the-arts methods.
vector was concatenated with four PHOG descriptors.
Alsuwaidi et al. [34] used a ground-breaking analytical
classification system in which they incorporated adaptive 3. MODELS AND METHODS
feature collection, novelty identification and ensemble
learning with the hyper spectral datasets. Singh et al. [35] This section discusses in detail about the proposed method
presented an automated approach to differentiate between of apple leaves disease classification and identification
Neem and Bakain using the texture characteristics of its leaves including dataset collection, pre-processing, segmentation,
and then they used tree classifier to separate them in separate feature extractions, training, testing etc. It is organized in
classes. Brahimi et al. [36], suggested the employment of different subsections.
saliency maps for simulation to perceive and decode the CNN
classification process. This process of simulation improves the 3.1 Data set
clarity of deep learning frameworks and offers further
knowledge about plant disease symptoms. The dataset is composed of four types of apple leaves
Yue et al. [37] proposed a super resolution model that relies downloaded from Kaggle [43]. These are healthy and
on the residual deep recursive network and provides the state- unhealthy apple leaves. Among the unhealthy apple leaves, it
of-the-arts performance as compared to the traditional consists of three types of infections namely: apple scab, black
methods. Iqbal et al. [38] presented a review on the different rot and apple rust. Figure 1 shows healthy and unhealthy
approaches of identification and classification of diseases sample of images in the dataset collection used in this article.
specific to citrus plants.
Dhingra et al. [39] has outlined a neutrosophic approach
based on computer vision for plant disease analysis. This
system uses a fuzzy set extension technique based on
neutrosophical logic segmentation to analyze the area of
interest and then, new feature subset is assessed on the basis
of segmented area to classify the basil leaf as healthy or (a) Healthy leaf samples
diseased. Picon et al. [40] used the Deep Residual Neural
Network-based algorithm to detect plant diseases under
specific acquisition circumstances where various adaptations
have been suggested for early disease discovery. This research
analyzes the early identification success of three related
European endemic diseases of wheat: Septoria, Tan Spot and
Rust. (b) Unhealthy leaf samples
Wu et al. [41] proposed a new solution to crop disease
detection focused on multi-functional sparse constrain system Figure 1. Dataset collection of apple leaf images (a) Healthy
that mainly includes three phases: segmentation of lesions, (b) Unhealthy
extraction of features and disease detection. They applied this

Figure 2. Flow diagram of basic system model for proposed approach

1095
3.2 Basic model design 3.4 Segmentation technique

Following are fundamental steps followed to design the Segmentation of images is an arduous process because of
proposed systems: the intricacy and variety of images [49]. Factors like lighting
1. Pre-processing of data: The main objective of pre- [50], contrast, interference etc. affect the outcome of
processing is to highlight the information which is concealed segmentation. The segmentation aim is to identify the areas of
by the contrast stretching [44] approach for improved contrast. concern to determine the disease. We have advocated FCM
2. Segmentation: The concerned area is identified after clustering approach for segmentation. FCM is a clustering
pre-processing using FCM clustering segmentation. technique that enables a piece of information to be a member
3. Extraction of features: The classification model used of more than one cluster and therefore, it belongs to the class
in the proposed work is convolutional neural networks [45] of soft segmentation technique. These techniques are
which is also used for extraction of features. popularly used for image segmentation since much better
4. Classification: The methodology used for the details from the main image could be retrieved in comparison
identification of leaf disease is CNN. to hard segmentation approaches. FCM is a technique of
clustering that allocate pixels to unlabelled clusters with
The steps mentioned above are addressed in more depth in different membership rates. FCM clustering segmentation
the subsequent parts. The flow diagram given in Figure 2 algorithm is outlined in algorithm 1 [51]. Unlike other
presents the proposed approach. The primary purpose of this clustering methods, where data point must exclusively belong
work is to enable the system to learn the characteristics that to one cluster center, in FCM data clustering, point is assigned
differentiate one class from another. In order to achieve this, membership to each cluster center as a result of which data
dataset can be increased by using augmented images to point may belong to more than one cluster center. Figure 4
improve the network’s probability of learning the correct shows the result on unhealthy leaf sample after applying pre-
features [46]. Figure 3 shows the result of augmentation on processing and segmentation steps.
unhealthy apple leaf images.

(a) (b) (c)

(a) (b)
Figure 4. Result of pre-processing and segmentation on
Figure 3. (a) Original (b) augmented samples of unhealthy unhealthy leaf sample, (a) Original sample, (b) Contrast
apple leaf stretched sample, (c) FCM segmented sample

3.3 Preprocessing Algorithm 1: FCM Clustering segmentation

1. Select principal centroids Ci.
Image preprocessing is an important step in examination 2. Use Eq. (2) to evaluate the disparity between the
and manipulation of a digital photo [47], particularly with a vectors and the centroid.
view to highlight problem regions i.e., diseased region after
data gathering is done. However, visual enhancement is one of
the complexities of image processing [48], and is also very 𝑑𝑖𝑠𝑝𝑎 = √(𝑑𝑥 − 𝑝𝑥 )2 + (𝑑𝑦 − 𝑝𝑦 )2 (2)
unique to a specific task. In this work, the contrast stretching
algorithm is used to enhance the image. The contrast where, dispa is Euclidean distance between data points/clusters
enhancement methods are used to extend the range of center (dx, dy) and (px, py).
brightness levels in an image, so that the image can be viewed 3. Evaluate the membership coefficient for vector
effectively in the way the analyst wishes. The contrast level in belonging to a particular cluster using Eq. (3).
an image can differ due to poor lighting or inappropriate
setting in the calibration device. 1
The method of contrast stretching is to examine the spread 1 𝑓−1
[ ]
of pixel concentrations in a picture and then dynamically 𝑑𝑖𝑠𝑝𝑏𝑎
𝑚𝑒𝑚𝑏 (𝑑𝑎 ) = 1 (3)
resize the picture so as to include all levels of intensity falling 1 𝑓−1
within 2nd and 98th percentiles. The formula for global ∑𝑁
𝑛=1 [𝑑𝑖𝑠𝑝 ]
𝑛𝑎
contrast stretching is governed by the following equation
where, f represents fuzziness coefficient. This value of
(𝑖𝑛𝑅𝐺𝐵 − 𝑚𝑖𝑛𝑅𝐺𝐵 ) membership has to be calculated for all the clusters; n is the
𝑜𝑢𝑡𝑅𝐺𝐵 (𝑥, 𝑦) = 255 ∗ [ ] (1) number of data points belonging to the cluster.
𝑚𝑖𝑛𝑅𝐺𝐵 − 𝑚𝑖𝑛𝑅𝐺𝐵
4. Determine new centroid using Eq. (4) and upgrade
membership coefficient from step 3.
where, inRGB(x, y) is the original RGB value of the pixel,
outRGB(x, y) is the new RGB value of the pixel, minRGB is
∑𝑎[𝑚𝑒𝑚𝑏 (𝑑𝑎 ]𝑓 𝑑𝑎
minimum value between the components, and maxRGB is 𝑃𝑏𝑛𝑒𝑤 = (4)
maximum value between the RGB components (red, green, ∑𝑎[𝑚𝑒𝑚𝑏 (𝑑𝑎 ]𝑓
and blue) of the original image.
5. If the centroids do not alter, stop. Else go to step 4.

1096
3.5 Feature extraction and classification Pooling is another crucial layer of CNN. It is a type of
nonlinear down sampling. The size of the feature maps
The remarkable improvement in performance, achieved for obtained from convolutional layers could be reduced using this
different tasks using deep neural networks inspired us to layer to accomplish spatial invariance. Completely connected
employ it for the purpose of image classification in the present layers are inserted before a CNN’s classification output, and
work. We used CNN to extract features and classify images. used before classification to straighten the result. The end-
The output of the segmentation is given to the CNN to classify product of the final pooling or convolutionary layer is the entry
whether an image is healthy or not. A CNN is a kind of feed- to the totally linked layer. Figure 5 represents the CNN model
forward network and is an end-to-end pipeline methodology parameters used in this work.
that can inevitably uncover the discriminatory characteristics
for image classification. Since the features in CNN are not
extracted and implemented on the basis of human knowledge, 4. EXPERIMENTAL OBSERVATION
they are very less prone to artificial feature extraction.
CNN consists of layers; however, these layers are not fully 4.1 Dataset of apple leaf images
interconnected. They have filtering process, which includes
series of cube like shape weights applied all through the The dataset collection consists of 400 images, out of which
picture. Each two-dimensional filtration portion is called 200 images are healthy leaf samples and the rest consists of
“kernel”. Major component of CNN is the convolutional layer. various categories of diseased leaf samples that includes apple
A convolution functions between the higher feature maps of scab, black rot, and apple rust. All the experiments are
the present layer and the convolution kernels decide the performed employing “Keras framework on top of tensorflow”.
resulting feature map for every convolution layer. The output The dataset is expanded by utilizing “ImageDataGenerator
characteristic map can be given by Eq. (5) [26]. class” of tensorflow.

𝐼𝑛𝑏𝑙 = ∑ 𝐼𝑛𝑎𝑙−1 ∗ 𝑐𝑎𝑏

𝑙
+ 𝐵𝑏𝑙 4.2 Evaluation metrices for system model
(5)
𝑎∈𝑋𝑏
The efficiency of the proposed system was assessed using
different evaluation parameters [52, 53]. A brief overview
where, l is the lth layer, Bb denotes bias, cab is the
about these metrices is provided below.
convolutional kernel and Xb is set of input feature maps. The
1. Accuracy(A): The accuracy metric is defined as the
neural model’s learning capacity is affected by the ReLU
ratio of the number of correctly classified examples to the total
activation function and it has a quick convergence speed. This
number of classified examples.
approach is therefore used for the output of each convolutional
layer. Mathematically it can be expressed as given in Eq. (6)
A = (TP + FN)/(TP + FP + TN + FN) (7)
[26]:
2. Precision(P): Precision is used to quantify the
𝑓(𝑧) = 𝑚𝑎𝑥𝑖𝑚𝑢𝑚(0, 𝑧) (6)
positive patterns correctly predicted in a positive class from
total predicted patterns.

P = TP/(TP + FP) (8)

3. Recall(R): Recall is used to calculate the proportion

of positive occurrences correctly categorized.

R = TP/(TP + TN) (9)

4. F1-score(F): F1-score or F-measure metric

represents the harmonic mean between recall and precision
values [52].

F = 2 ∗ P ∗ R/(P + R) (10)

where, TP = True Positive, FP = False Positive, TN = True

Negative and FN = False Negative.

4.3 System model evaluation (Training-Validation

observation)

In this section, based on the training-testing dataset listed

within Table 2 and Table 3, we examined the predictive
outcome of CNN model. Data distribution for this model is set
at 80% and 20% respectively in the training-validation
Figure 5. Parameters of CNN model analysis.

1097
 Table 2. Data partitioning of original image samples for
training and validation

Total Training Validation

Data set
samples samples samples
Healthy 200 158 42
Unhealthy 200 162 38

Table 3. Data partitioning of augmented image samples for

training and validation

Augmented Total Training Validation

Dataset samples samples samples
Healthy 952 757 195 Figure 9. Comparison of CNN model on adding each
Unhealthy 944 759 185 proposed step (in terms of F1-score)

Figures 6-9 demonstrates the CNN model’s classification

accuracy by adding each proposed step with respect to
different assessment parameters, defined in section 4.2. Since
on training the model with bigger dataset, we can expect the
outcome to be more accurate on test data. Therefore, the
assessment parameters are also evaluated on the augmented
dataset. Augmented dataset is the improved version of existing
dataset in terms of its size and variety, without the need to
explicitly gather new data. It is clear from Figure 6 that the
proposed model achieves 94% accuracy with lower dataset
and this has been further improved to 98% with the use of
augmented dataset.

Figure 6. Comparison of CNN model on adding each 4.4 Analysis on various Training-Validation partitions
proposed step (in terms of accuracy)
In addition, the accuracy is measured on various training-
validation divisions i.e. 50-50, 60-40, 70-30, 80-20 and 90-10,
as shown in Tables 4 and 5 for original and augmented dataset
respectively.

Table 4. Performance analysis on various training-validation

divisions in the original dataset

Training and Training Validation Accuracy

validation partitions samples samples rate
50-50% 200 200 85%
60-40% 240 160 88%
70-30% 280 120 89%
80-20% 320 80 94%
90-10% 360 40 95%
Figure 7. Comparison of CNN model on adding each
proposed step (in terms of precision) Table 5. Performance analysis on various training-validation
divisions in the augmented dataset

Training and Training Validation Accuracy

validation partitions samples samples rate
50-50% 948 948 91%
60-40% 1137 759 94%
70-30% 1327 569 95%
80-20% 1516 380 97%
90-10% 1706 190 98%

Accuracy rate is evaluated on both original as well as

augmented dataset to check the robustness of the proposed
model. The accuracy of CNN model depends strongly on its
training dataset size. But the presented system achieves
sufficiently high accuracy even with the small training dataset.
Figure 8. Comparison of CNN model on adding each
Results in Table 4 show that the presented method can achieve
proposed step (in terms of recall)
accuracy of 95% when 360 training and 40 validation image

1098
samples were used. Also, it is evident from Table 5 that with image processing techniques. IOSR Journal of Computer
an improved data set and higher training ratio, the proposed Engineering, 16(1): 10-16. https://doi.org/10.9790/0661-
system can achieve accuracy of up to 98%. 16151016
[2] Kulkarni, A.H., R.K., Ashwin Patil. (2012). Applying
4.5 Comparison with the existing systems image processing technique to detect plant diseases.
International Journal of Modern Engineering Research,
In this experiment, we measure our proposed model's 2(5): 3661-3664.
classification accuracy against the conventional methods [26, [3] Arsenovic, M., Karanovic, M., Sladojevic, S., Anderla,
28]. As outlined in Table 6, the proposed approach provides A., Stefanovic, D. (2019). Solving current limitation of
superior performance than other methods with far less deep learning based approaches for plant disease
augmented dataset size. The comparison with these systems is detection. Symmetry, 11(7): 939.
made on the grounds of common use of augmented dataset and https://doi.org/10.3390/sym11070939
to show how the proposed model performs better even the [4] Mainkar, P.M., Ghorpade, S., Adawadkar, M. (2015).
augmented set is much lower. The main idea of Liu et al. [26] Plant leaf disease detection and classification using
is to generate ample pathological representation of apple image processing techniques. International Journal of
images and build an innovative model of AlexNet based deep Innovative and Emerging Research in Engineering, 2(4):
CNN for disease identification. Waheed et al. [28] also 139-144.
achieved good results by collecting more dataset and tuning of [5] Fang, Y., Ramasamy, R.P. (2015). Current and
system variables. Waheed et al. [42] proposed optimized prospective methods for plant disease detection.
DenseNet model corn leaf and achieved an accuracy of Biosensors, 5(3): 537-561.
98.06%. In this work, the prime motive was to enhance the https://doi.org/10.3390/bios5030537
system accuracy with minimal dataset which is acquired by [6] Hyson, D.A. (2011). A comprehensive review of apples
using contrast stretching as pre-processing method and FCM and apple components and their relationship to human
clustering algorithm for segmentation. health. Advances in Nutrition, 2(5): 408-420.
https://doi.org/10.3945/an.111.000513
Table 6. Comparison of proposed model with the existing [7] Koutsos, A., Tuohy, K.M., Lovegrove, J.A. (2015).
models Apples and cardiovascular health - Is the gut microbiota
a core consideration? Nutrients, 7: 3959-3998.
Liu Hanson Waheed Proposed https://doi.org/10.3390/nu7063959
[26] [28] [42] method [8] Tadeusiewicz, R., Tylek, P., Adamczyk, F., Kielbasa, P.,
Original dataset Jablonski, M., Pawlik, P., Pilat, A., Walczyk, J.,
1053 >200 12,332 400
size Szczepaniak, J., Juliszewski, T., Szaroleta, M. (2017).
Augmented dataset Not
13689 33469 1896 Automation of the acorn scarification process as a
size available
Accuracy rate 97.62% 95% 98.06% 98%
contribution to sustainable forest management, case
study: Common oak. Sustainability, 9(12): 2276.
https://doi.org/10.3390/su9122276
[9] Barbedo, J.G.A. (2013). Digital image processing
5. CONCLUSION
techniques for detecting, quantifying and classifying
In this work, we proposed a novel approach to efficiently plant diseases. SpringerPlus, 2(1): 1-12.
use contrast stretching based preprocessing and fuzzy c-means https://doi.org/10.1186/2193-1801-2-660
segmentation, together with CNN to identify disease in apple [10] Dhingra, G., Kumar, V., Joshi, H.D. (2018). Study of
digital image processing techniques for leaf disease
leaves. The entire operation was outlined from collecting
detection and classification. Multimedia Tools Appl,
images to segmentation and eventually, feature extraction and
classification by CNN. Based on the results of preprocessing 77(15): 19951-20000. https://doi.org/10.1007/s11042-
followed by segmentation, a new deep convolutional neural 017-5445-8
network model has been developed that discover distinctive [11] Ghaiwat, S.N., Arora, P. (2014). Detection and
features automatically, and also determine apple leaf diseases classification of plant leaf diseases using image
processing techniques: A review. Int J Recent Adv Eng
accurately. The study proposed has been compared with
Technol, 2(3): 2347-2812.
existing state-of-the-art and the results were found quite
impressive. The developed system works better with 98% [12] Dhaygude, S.B., Kumbhar, N.P. (2013). Agricultural
accuracy rate and that too, with much lesser dataset size. plant leaf disease detection using image processing. Int J
Future research could focus on extending suggested work to Adv Res Electr Electron Instrum Eng, 2(1).
characterize each category of diseases separately and [13] Selvaraj, A., Shebiah, N., Ananthi, S., Varthini, S.V.
(2013). Detection of unhealthy region of plant leaves and
approximate the seriousness of the diseases identified. An
undiscovered combination of the extraction, collection of classification of plant leaf diseases using texture features.
features and learning approaches can also be analysed to Agric Eng Int CIGR, 15(1): 211-217.
improve the effectiveness of disease diagnosis and [14] Pujari, J.D., Yakkundimath, R., Byadgi, A.S. (2013).
identification models. Classification of fungal disease symptoms affected on
cereals using color texture features. Int. J. Signal Process.
6(6): 321-330.
REFERENCES [15] Vishnu, S., Ram, A.R. (2015). Plant disease detection
using leaf pattern: A review. International Journal of
[1] Gavhale, M.K.R., Gawande, U. (2014). An overview of Innovative Science, Engineering and Technology, 2(6):
the research on plant leaves disease detection using 774-780.

1099
[16] Muthukannan, K., Latha, P. (2015). A PSO model for Science and Computing, 7(3): 5324-5328.
disease pattern detection on leaf surfaces. Image [29] Yao, Q., Chen, G.T., Wang, Z., Zhang, C., Yang, B.J.,
Analysis and Stereology, 34(3): 209-216. Tang, J. (2017). Automated detection and identification
https://doi.org/10.5566/ias.1227 of white-backed planthoppers in paddy fields using
[17] Chung, C.L., Huang, K.J., Chen, S.Y., Lai, M.H., Chen, image processing. Journal of Integrative Agriculture,
Y.C., Kuo, Y.F. (2016). Detecting Bakanae disease in 16(7): 1547-1557. https://doi.org/10.1016/S2095-
rice seedlings by machine vision. Computers and 3119(16)61497-1
Electronics in Agric, 121: 404-411. [30] Sethy, P.K., Negi, B., Barpanda, N.K., Behera, S.K.,
https://doi.org/10.1016/j.compag.2016.01.008 Rath, A.K. (2017). Measurement of disease severity of
[18] Zhang, S.W., Wang, Z. (2016). Cucumber disease rice crop using machine learning and computational
recognition based on Global-Local Singular value intelligence. Cognitive Science and Artificial
decomposition. Neurocomputing, 205: 341-348. Intelligence, Springer, Singapore, pp 1-11.
https://doi.org/10.1016/j.neucom.2016.04.034 https://doi.org/10.1007/978-981-10-6698-6_1
[19] Ashourloo, D., Aghighi, H., Matkan, A.A., Mobasheri, [31] Fuentes, A.F., Yoon, S., Lee, J., Park, D.S. (2018). High-
M.R., Rad, A.M. (2016). An investigation into machine performance deep neural network based tomato plant
learning regression techniques for leaf rust disease diseases and pests diagnosis system with refinement
detection using hyper spectral measurement. IEEE filter bank. Frontiers in Plant Science, 9: 1162.
Journal of Selected Topics in Applied Earth https://doi.org/10.3389/fpls.2018.01162
Observations and Remote Sensing, 9(9): 4344-4351. [32] Sunny, S., Gandhi, M.P.I. (2018). An efficient citrus
https://doi.org/10.1109/JSTARS.2016.2575360 canker detection method based on contrast limited
[20] Ali, H., Lali, M.I., Nawaz, M.Z., Sharif, M., Saleem, B.A. adaptive histogram equalization enhancement.
(2017). Symptom based auto-mated detection of citrus International Journal of Applied Engineering Research,
diseases using color histogram and textural descriptors. 13(1): 809-815.
Computers and Electronics in Agric, 138: 92-104. [33] Zhang, S., Wang, H., Huang, W., You, Z. (2018). Plant
https://doi.org/10.1016/j.compag.2017.04.008 diseased leaf segmentation and recognition by fusion of
[21] Kaur, P., Singla, S., Singh, S. (2017). Detection and superpixel K-means and PHOG. Optik, 157: 866-872.
classification of leaf diseases using integrated approach https://doi.org/10.1016/j.ijleo.2017.11.190
of support vector machine and particle swarm [34] Alsuwaidi, A., Grieve, B., Yin, H. (2018). Feature-
optimization. Int. J. Adv. Appl. Sci., 4(8): 79-83. ensemble based novelty detection for analyzing plant
[22] Ma, J.C., Du, K.M., Zhang, L.X., Zheng, F.X., Chu, J.X., hyperspectral data. IEEE Journal of Selected Topics in
Sun, Z.F. (2017). A segmentation method for greenhouse Applied Earth Observations and Remote Sensing, 11(4):
vegetable foliar disease spots images using color 1041-1055.
information and region growing. Computers and https://doi.org/10.1109/JSTARS.2017.2788426
Electronics in Agriculture, 142: 110-117. [35] Singh, N.P., Nagahama, T., Yadav, P., Yadav, D. (2018).
https://doi.org/10.1016/j.compag.2017.08.023 Feature based leaf identification. 5th IEEE Uttar Pradesh
[23] Mondal, D., Kole, D.K., Roy, K. (2017). Gradation of Section International Conference on Electrical,
yellow mosaic virus disease of okra and bitter gourd Electronics and Computer Engineering (UPCON-2018),
based on entropy based binning and Naive Bayes pp. 1159-1165.
classifier after identification of leaves. Computers and [36] Brahimi, M., Arsenovic, M., Laraba, S., Sladojevic, S.,
Electronics in Agriculture, 142: 485-493. Boukhalfa, K., Moussaoui, A. (2018). Deep learning for
https://doi.org/10.1016/j.compag.2017.11.024 plant diseases: detection and saliency map visualization.
[24] Zhang, C.L., Zhang, S.W., Yang, J.C., Shi, Y.C., Chen, Human and Machine Learning Human–Computer
J. (2017). Apple leaf disease identification using genetic Interaction Series, pp. 93-117.
algorithm and correlation based feature selection method. [37] Yue, Y., Cheng, X., Zhang, D., Wu, Y.Z., Zhao, Y., Chen,
International Journal of Agricultural and Biological Y.Q., Fan, G.H., Zhang, Y.H. (2018). Deep recursive
Engineering, 10(2): 74-83. super resolution network with Laplacian Pyramid for
https://doi.org/10.3965/j.ijabe.20171002.2166 better agricultural pest surveillance and detection.
[25] Singh, V., Mishra, A.K. (2017). Detection of plant leaf Computers and Electronics in Agriculture, 150: 26-32.
diseases using image segmentation and soft computing https://doi.org/10.1016/j.compag.2018.04.004
techniques. Information Processing in Agriculture, 4(1): [38] Iqbal, Z., Khan, M.A., Sharif, M., Shah, J.H., Rehman,
41-49. https://doi.org/10.1016/j.inpa.2016.10.005 M.H., Javed, K. (2018). An automated detection and
[26] Liu, B., Zhang, Y., He, D.J., Li, Y. (2017). Identification classification of citrus plant diseases using image
of apple leaf diseases based on deep convolutional neural processing techniques: A review. Computers and
networks. Symmetry, 10(1): 11. Electronics in Agriculture, 153: 12-32.
https://doi.org/10.3390/sym10010011 https://doi.org/10.1016/j.compag.2018.07.032
[27] Dechant, C., Wiesner-Hanks, T., Chen, S., Stewart, E.L., [39] Dhingra, G., Kumar, V., Joshi, H.T. (2019). A novel
Yosinski, J., Gore, M.A., Nelson, R.J., Lipson, H. (2017). computer vision based neutrosophic approach for leaf
Automated identification of northern leaf blight-infected disease identification and classification. Measurement,
maize plants from field imagery using deep learning. 135: 782-794.
Phytopathology, 107(11): 1426-1432. https://doi.org/10.1016/j.measurement.2018.12.027
https://doi.org/10.1094/PHYTO-11-16-0417-R [40] Picon, A., Alvarez-Gila, A.., Seitz, M., Ortiz-Barredo, A.,
[28] Hanson, A.J., Joy, A., Francis, J.N. (2017). Plant leaf Echazarra., J., Johannes, A. (2019). Deep convolutional
disease detection using deep learning and convolutional neural networks for mobile capture device-based crop
neural network. International Journal of Engineering disease classification in the wild. Computers and

1100
 Electronics in Agriculture, 161: 280- 290. processing techniques - A survey. International
https://doi.org/10.1016/j.compag.2018.04.002 Multidisciplinary Research Journal, 5(11).
[41] Wu, Y.R., Li, J.H. (2019). Multi-feature sparse constrain [48] Shukla, K.N., Potnis, A., Dwivedi, P. (2017). A review
model for crop disease recognition. Applied Ecology and on image enhancement techniques. International Journal
Environmental Research, 17(4): 9229-9245. of Engineering and Applied Computer Science (IJEACS),
https://doi.org/10.15666/aeer/1704_92299245 2(7): 232-235. https://doi.org/10.24032/ijeacs/0207/05
[42] Waheed, A., Goyal, M., Gupta, D., Khanna, A., [49] Wang, Z.B., Wang, K.Y., Yang, F., Pan, S.H., Han. Y.Y.
Hassanien, A.E., Pandey, H.M. (2020). An optimized (2018). Image segmentation of overlapping leaves based
dense convolutional neural network model for disease on Chan-Vese model and Sobel operator. Information
recognition and classification in corn leaf. Computers Processing in Agriculture, 5(1): 1-10.
and Electronics in Agriculture, 175: 105456. https://doi.org/10.1016/j.inpa.2017.09.005
https://doi.org/10.1016/j.compag.2020.105456 [50] Pratiwi, D., Kartowisastro, I.H. (2015). Object
[43] Bhattarai, S. New Plant Diseases Dataset segmentation under varying illumination effects.
https://www.kaggle.com/vipoooool/new-plant-diseases- NewTrends in Intelligent Information and Database
dataset, accessed on 12 July 2020. Systems Studies in Computational Intelligence, pp. 13-
[44] Perumal, S., Velmurugan, T. (2018). Preprocessing by 21. https://doi.org/10.1007/978-3-319-16211-9_2
contrast enhancement techniques for medical images. [51] Kumari, A., Meenakshi, S., Abinaya, S. (2018). Plant
International Journal of Pure and Applied Mathematics, leaf disease detection using fuzzy c-means clustering
118(18): 3681-3688. algorithm. International Journal of Engineering Research
[45] Bodapati, J.D., Veeranjaneyulu, N. (2019). Feature and Technology, 6(3): 157-163.
extraction and classification using deep convolutional [52] Hossin, M., Sulaiman, M.N. (2015). A review on
neural networks. Journal of Cyber Security and Mobility, evaluation metrices for data classification evaluations.
8(2): 261-276. https://doi.org/10.13052/jcsm2245- International Journal of Data Mining and Knowledge
1439.825 Management Process, 5(2): 1-11.
[46] Parente, A.P., de Souza, M.B., Valdman, A., Folly, https://doi.org/10.5121/ijdkp.2015.5201
R.O.M. (2019). Data augmentation applied to machine [53] Novakovic, J.D., Veljovic, A., Ilic, S.S., Papic, Z.,
learning-based monitoring of a pulp and paper process. Tomovic, M. (2017). Evaluation of classification models
Processes, 7(12): 958. in machine learning. Theory and Applications of
https://doi.org/10.3390/pr7120958 Mathematics and Computer Science, 7(1): 39-46.
[47] Muthuselvi, S., Prabhu, P. (2016). Digital image

1101