Technological

Communication
Biosc.Biotech.Res.Comm Vol 12 (4) 1081-1091 (2019)

An Identification and Detection Process for

Leaves Disease of Wheat Using Advance
Machine Learning Techniques
Nafees Akhter Farooqui* and Ritika
Department of Computer Science, DIT University Dehradun-248009, Uttarakhand, India

ABSTRACT
In India, the economic, political, and social stability depends directly as well as indirectly on agricultural
productivity. The agricultural product quality and productivity are affected by the different diseases in plants.
Therefore, diseases detection in plants is very important in the agriculture field. Leaf disease detection by
using different machine learning techniques is a very popular field of study. There have already been many
promising outcomes but only a few real-life applications that can make life simpler for the farmer. Machine
learning technique becomes the most accurate and precise paradigm for the detection of plant’s disease which is
helpful to reduce an oversized work of watching in huge farms of crops, and an early stage itself it detects the
symptoms of diseases on plant leaves. In this paper, we use a novel approach for detection and identification
of leaf diseases by K-means clustering, multi-class SVM, and advanced neural network techniques to process
data, on plant disease detection. Gray Level Co-occurrence Matrix (GLCM) is used for feature extraction. The
disease classification is done using SVM classifier, and the detection accuracy is improved by optimizing the
data using the Advanced Neural Network (NN). The extracted features are fit into the network. This is a great
success, demonstrating the feasibility of this approach in plant disease diagnosis and high crop yielding. The
aim of our research is solving the problem of detection and preventing diseases of crops. We also demonstrate
the model for leaf diseases detection system for future trends of hyperspectral data.

KEY WORDS: ADVANCED NEURAL NETWORK, GRAY LEVEL CO-OCCURRENCE MATRIX, HYPERSPECTRAL DATA,
K-MEANS CLUSTERING, SUPPORT VECTOR MACHINE

ARTICLE INFORMATION
*Corresponding Author: [email protected]
Received 05th Oct 2019
Accepted after revision 30th Nov 2019
Print ISSN: 0974-6455 Online ISSN: 2321-4007
CODEN: BBRCBA

Thomson Reuters ISI Web of Science Clarivate Analytics

USA and Crossref Indexed Journal

NAAS Journal Score 2019 (4.38) SJIF: 2019 (4.196)

A Society of Science and Nature Publication, Bhopal India
2019. All rights reserved.
Online Contents Available at: http//www.bbrc.in/
DOI: 10.21786/bbrc/12.4/31
1081
 Farooqui & Ritika

INTRODUCTION pen-paper calculation into the reality of high

magnitude. Therefore, this work can describe
In the global economy, agriculture plays a crucial the finding of a solution for minimizing the cost
role. With the continued expansion of the human by avoiding manual monitoring and an expert
population, the pressure on the agricultural system requirement for automatic detection of leaf
will increase. Agri-technology and precision diseases in a large area (LeCun. et al., 2015).
farming, now also known as digital agriculture,
have emerged as fresh scientific fields using The plants are vulnerable to various disease-
data-intensive methods to boost agricultural caused illnesses and assaults. The impacts on
productivity while minimizing its effect on the crops can be characterized by several factors,
environment. The Advance Machine Learning such as temperature, humidity, dietary excess or
(ML) method offers high-performance computing loss, bacterial, viral, and fungal illnesses. These
to generate fresh possibilities in agricultural diseases can demonstrate different physical
operating settings to unravel, quantify, and characteristics on the leaves together with the
comprehend data-intensive procedures. ML is plants, such as modifications in form, colors, etc.
defined as the scientific field that gives machines Due to various comparable patterns, the above
the ability to learn without strict programming, modifications are hard to distinguish, making
among other definitions. Wheat is the largest their identification, but the early detection and
humanoid nutrition in the world that can be fed prevention will prevent the diseases in the entire
directly from any other crop. Since nutrition is crop. We will use K-means clustering, multi-class
becoming increasingly essential in Asia, where SVM, and advanced neural network techniques on
most of the poor people reside (Shuang-Qi et al., plant disease detection. Gray Level Co-occurrence
2017). Agricultural productivity is the source of Matrix (GLCM) is used for feature extraction
economic growth for developing countries like of plant leaf diseases that affect various plants
India. (Vidyashanakara et al., 2018).

Chronologically, the primary goal of farming Here we take some of the articles concerning
is to provide the country with food and feed the identification of plant leaf illnesses using
it. Plant diseases not only pose a global threat different sophisticated methods and some of them
to food security but can also have disastrous shown below. (Fujita et al.,2016) recommended
consequences for smallholder farmers who the CNN classifier for cucumber diseases. It
depend on healthy crops for their livelihoods. uses two datasets for training and validation.
There are more than 80 percent crop production In addition to the healthy class, these datasets
generated by small farmers in the developing comprise seven distinct kinds of disease. The first
countries (UNEP, 2013), and accounts of yield dataset is made up of 7320 targeted images of
losses of more than 50 percent are prevalent captured leaves under excellent circumstances.
owing to pests and diseases (Abebe et.al.,2017). The second dataset comprises of 7520 images
Also, the biggest proportion of hungry individuals captured that are healthy and unhealthy, Lu
(50%) live in smallholder farming homes (Sanchez et al.(2017) defined an automatic diagnostic
et.al, 2005), making smallholder farmers a group system for wheat disease based on a supervised
especially susceptible to pathogen-derived food deep learning framework that integrates wheat
supply disruptions. So, these leaf diseases in disease identification and disease region. In
any forms in crop tend to cause a reduction in addition, a fresh wheat disease image dataset,
quality, yield, and fiscal progression, respectively. Wheat Disease Database 2017 (WDD, 2017), is
Therefore, looking to the current farming arena, being gathered to confirm the system's efficiency.
instead of watching the crop through the naked Kamilaris et al.(2018) discussed and studied 40
eyes by a designated specialist where it does research papers using deep learning methods
need a lot of efforts to implicate. Hence, as a that addressed multiple problems in agricultural
result, the automation essentiality of leaf disease and food production. Also, examined the specific
identification and its management has turned the agricultural problems with the help of the specific

BIOSCIENCE BIOTECHNOLOGY RESEARCH COMMUNICATIONS An Identification and Detection Process for Leaves 1082
Farooqui & Ritika

models and frameworks. Ferentinos et al. (2018) management, (Bannerjee et al.,2018). There are
explained convolutional neural network models various methods had been already developed to
to detect and diagnose crop disease using plain prevent the losses of the leaf disease. However,
leaf images of healthy and diseased crops using for many farmers, these methods are unavailable
deep learning methods. The training model was and require thorough knowledge of the domain or
carried out using an open database of 87,848 a lot of money and resources to be implemented.
images, consisting of 25 distinct crops in a In this field of research, there are several existing
set of 58 different classes, including healthy limitations. One of them is that currently available
plants. Several model architectures were trained data sets do not include images from real-life
with the greatest outcomes reaching 99.53 situations that have been compiled and labelled.
percent achievement rate when defining the Another limitation is that the existing methods
corresponding combination. Kulkarni et al.(2012) accuracy is not good for the multiple disease in
described a methodology for early and accurate same plant, (Arsenovic et al.,2019). Therefore,
detection of plant diseases using the artificial in this paper our agenda is to resolve the above
neural network (ANN) and various techniques for limitations by proposing the architectural model
image processing. Since the suggested method is for disease detection in wheat leaf by combination
based on the classification ANN classifier and the of two methods and compare their accuracy.
extraction function Gabor filter, it delivers better
outcomes with a recognition rate of up to 91%. Leaf Disease Learning Model From Image: We are
proposing a Leaf Disease Learning Model from
In Malus domestica Bashir et al (2012) provided Image data, processing begins in this scheme
disease detection using an efficient technique with data collection in the form of the knowledge
such as K-mean clustering, texture, and color base, then training the data, which is used to
examination. It utilizes the texture and color apply the different machine learning algorithm.
characteristics that usually appear in ordinary Feature extractor steps to be permitted through
and impacted fields to classify and acknowledge pre-processing and then identify the illnesses by
distinct agriculture. Kaundal et al.,(2006) implementing the various algorithms. Finally, it is
contrasted the results of standard multiple possible to classify the defective region from the
regression, neural network, and support vector images. Fig.1 shows our proposed architecture.
machine (SVM). It was discovered that the
SVM-based regression approach gives better material and methods
result. The best description is the association of
environmental circumstances with the amount The proposed methodology for disease
of disease that could be helpful in disease identification is explained in the following

Figure 1. Architecture of Wheat Leaf Disease Detection System.

1083 An Identification and Detection Process for Leaves BIOSCIENCE BIOTECHNOLOGY RESEARCH COMMUNICATIONS
 Farooqui & Ritika

sub-sections Dataset Preparation: We can circumstances of agriculture. Fig. 2 shows

download images from the website https:/ random samples of images from different wheat
plantvillage.psu.edu/ using the plant and disease classes. The complexity of the images is evident,
names of the keywords. All images can be contributing to several elements such as various
subsequently categorized into distinct groups. It leaves and other plant components, irrelevant
is an open database with approx. 14,308 Healthy items, distinct ground textures, shading impacts,
and infected plant leaf pictures were used to etc. Initially, the entire database was split into
train and test the leaf disease detection system. two datasets, the training set, and the test set,
The database used here involves five distinct by dividing the 14,308 images randomly so that
classes, where each class is described as a couple 75% of them formed the training set, and 25%
of leaves of crops and associated disease, while formed the test set. Thus, 10,731 images were used
some classes contain healthy crops. There are 30 for the model training, while the remaining 3,577
distinct healthy or diseased plants in these five images were kept testing the model performance
groups. More than 1/3 of the images accessible in the classification of new, previously "unseen"
(33.3%) were recorded in the field under actual images.

The first step for Leaf Disease Detection System

Figure 2. Random samples of images from
initialization. First, take the wheat leaf image
different Wheat Leaf Disease and Healthy Classes
that has captured from the digital camera. The
(A) Healthy Leaf, (B) Brown Rust Effected Leaf,
color of the input image is primary. The primary
(C) Black Chaff Effected Leaf, (D) Powdery Mildew
color is then transformed as required into a gray-
Effected Leaf, (E) Yellow Rust Effected Leaf.
scaled color. Then we experimented with a gray-
scaled version of the Plant Village dataset and
subsequently run all the tests on a Plant Village
dataset version where the leaves were segmented,
removed all additional background information
that could bring some inherent information. We
have evaluated the color, lightness, and saturation
aspects of the multiple components of the images.
One of the processing measures also enabled us
Figure 3. Sample images used in separate to readily be solved color casts, which occur to
experimental settings from the three distinct be very powerful in some of the dataset subsets.
variants of the Plant Village data set. (A) Leaf1 Fig.3 shows for a randomly selected set of leaves
color, (A1) Leaf1 grayscale, (A2) Leaf1 segmented, with the different versions of the same leaf.
(B) Leaf2 color, (B1) Leaf2 grayscale, (B2) Leaf2
segmented, (C) Leaf3 color, (C1) Leaf3 grayscale, Image preprocessing: Images can be in various
(C2) Leaf3 segmented. formats, quality and resolution in the dataset. For
example, images with lower resolution and less
than 500px dimensions will not be considered
as valid images for the dataset. To extract the
necessary data more effectively from the image
using pre-processing methods for image resizing
and contrast improvement.

Feature Extraction: Machine Learning techniques

contain an element at the designing stage. In
this stage, specialists propose the hand-made
highlights to encourage gaining from precedents.
This stage is significant and influences the general
execution of the learning framework. Shockingly,

BIOSCIENCE BIOTECHNOLOGY RESEARCH COMMUNICATIONS An Identification and Detection Process for Leaves 1084
Farooqui & Ritika

include designing is a manual part of the machine relatives can be estimated in Table1. The graphical
learning technique, and it takes more time (Le Cun. representation of the extracted feature is shown
et al., 2015). On the other hand, in an advanced in Fig 4.
machine learning technique feature extraction
is embedded with the learning algorithm where Training: In this step, train the image of the wheat
features are extracted in a fully automated way leaf by using K-means Clustering, Support Vector
without any intervention by a human expert. Machine, and advance neural network for making
Feature extraction plays a vital role in the process an image classification model.
of facsimile cataloging. Hence, GLCM could be
an effectual and right resourceful technique for Testing: The test set for leaf prediction as healthy/
statistical parameter extraction based on texture unhealthy with its disease name will be used in
mining (Gebejes et. al, 2013). this stage to assess the classifier's output.

The image features include Correlation, Entropy, Approaches K-Means Based Clustering Method:
Variance, Homogeneity, Contrast, Energy, and This stage involves the use of the K-Means
Mean are computed. The resulted topographies algorithm to cluster an image. Using a group of
using the monochrome concentration and K-classes, it is a very helpful technique for entity
positioning can be indicated by the association recognition (Dubey et al., 2013). It can be done by
of active dotted particles whose numerical limiting the square summation distance between

Table 1. Average Extracted Features Vector Using GLCM Method.

S. Contr Corre Energy Homog Mean Standard Ent RMS Variance Kurtosis Skew
No. ast lation eneity Deviation ropy ness

1 0.520 0.82726 0.81 0.956992 11.42 42.14 1.088 3.194 1523.1 17.921 1.8645
2 0.453 0.88268 0.741 0.942985 18.62 51.96 1.6572 4.679 2206.1 10.582 2.3426
3 1.195 0.76266 0.57 0.931926 25.981 60.49 2.1984 7.044 3429.7 7.2729 1.8866
4 0.6984 0.91896 0.602 0.955868 35.465 71.82 2.391 7.347 4946.8 4.3601 2.3403
5 1.2426 0.76266 0.78 0.939879 12.649 46.03 1.0954 3.338 1823.7 16.636 3.0963
6 1.2304 0.91896 0.376 0.932311 36.484 61.36 3.7752 9.044 3009.9 5.777 2.3511
7 0.7108 0.68088 0.655 0.929282 24.433 58.77 2.0447 4.267 2085.4 7.2405 3.2904
8 0.9014 0.80841 0.576 0.927381 35.799 71.94 3.6071 7.978 3948 5.0127 3.1041
9 1.1222 0.87418 0.698 0.943924 28.194 69.98 2.4453 6.015 3597.8 6.7964 3.0347
10 1.361 0.90097 0.724 0.935854 18.952 55.79 1.9201 5.197 2359.1 11.386 4.1762
11 0.7258 0.8651 0.614 0.925867 23.879 56 2.135 5.346 2258.1 7.4625 3.7641
12 0.4065 0.73879 0.75 0.94786 17.785 56.2 1.5335 4.562 2827.8 12.644 0.8359
13 1.221 0.8602 0.734 0.923988 15.923 48.57 1.4202 4.269 2039.3 11.61158 1.6122
14 0.7435 0.91903 0.757 0.938948 18.945 56.55 1.8782 4.812 2798.3 10.949 2.836
15 0.2743 0.70682 0.823 0.976112 13.225 44.84 2.1005 5.159 1703.5 20.069 1.3342
16 0.2681 0.85746 0.786 0.962688 14.843 47.85 2.0682 4.844 1490.1 16.601 1.3215
17 0.6847 0.91649 0.378 0.906731 59.88 78.53 4.5886 9.207 2859.6 2.1288 4.221
18 1.1217 0.92901 0.48 0.896516 40.152 73.24 3.153 7.711 4661.7 4.0614 3.0734
19 0.5468 0.92833 0.723 0.947347 17.922 49.77 1.6294 4.534 2051.4 10.111 0.4327
20 1.8418 0.87307 0.478 0.883095 45.042 77.61 2.9739 7.068 4540.2 3.109 0.745
21 1.5751 0.84231 0.843 0.903862 47.808 80.98 3.1664 7.936 5354.2 3.0829 1.8645
22 0.8019 0.81905 0.709 0.96402 12.901 46.13 2.0809 5.212 2015.2 19.969 2.3426
23 0.328 0.86446 0.237 0.944129 16.123 44.55 1.6311 2.662 352.88 12.241 1.8866
24 1.4987 0.76745 0.27 0.877284 77.399 84.8 4.79162711.2647 6629.51 1.5772 2.3406
25 1.346 0.86944 0.458 0.912759 62.05059 74.62 4.5629 10.71 4388.6 2.1224 3.0963
26 0.5789 0.87351 0.789 0.97456 68.999 66.78 4.3245 6.745 345.89 2.0789 2.0098
27 1.673 0.86979 0.399 0.982114 38.987 56.93 1.89789 3.679 661.78 9.0789 1.789
28 0.456 0.93412 0.897 0.911239 47.345 33.78 3.4567 5.8769 2897.9 13.1345 0.3344
29 1.876 0.8569 0.769 0.901987 59.8796 77.84 2.9897 10.78 2345.67 2.45908 5.789
30 1.798 0.789 0.749 0.80796 53.5346 79.1 1.79854 4.8999 3987.9 10.9089 1.0459

1085 An Identification and Detection Process for Leaves BIOSCIENCE BIOTECHNOLOGY RESEARCH COMMUNICATIONS
 Farooqui & Ritika

the equivalent cluster and an entity to find the This process has already been implemented in
thought-provoking fragment of the source image. other leaf images to cluster more than one groups
In K-Means clustering techniques, clusters are having the respective diseases (Sethy et al., 2017).
determined by pixel groups with the same value After the successful implementation K-Means
current in an image. Practically, this new image cluster-based segmentation, the percentage of
processing technique's computational acceleration the infected area calculated, and features are
is very quick and provides a more precise output. extracted.
The input data set is divided into K set of clusters,
and each cluster is represented by a completely Classification of Diseases by SVM: In this phase,
variable cluster center. Initially, known values are the classification and comparison for the leaf
referred to as seed points and data points are also diseases of the wheat crop have done through by
known as inputs. Only the K-Means clustering storing the corresponding feature set of values to
method can be used to estimate the distances their respective dataset. Here a fully controlled
between the centers, inputs, and allocate inputs learning method of classification is used that is
to the closest center. Following are the steps for none other than support vector machines (SVM)
K-means grouping: with high dimension spaces, efficient memory,
and versatile decision function. Generally, SVM
Step 1: Image confirmation is categorized into two types: Linear SVM and
Step 2: Image transformation from primary color Multiclass SVM. Linear SVM is used to classify
space to L*a*b* Color Space, which helps in the two kinds of data set, and multiclass SVM is used
findings of pictorial transformations that present to classify more than two kind of data set. So
in the primary color space. multiclass SVM is used to classify four types of
wheat leaf diseases. First, the extracted feature
Step 3: Here the colors classification can be done dataset is optimized using the optimization
in 'a*b*' Space using K-means. technique, and then multiclass SVM is used
for the classification process. The quantity of
Step 4: From the results, pixel ordering, and cataloging is performed by the proportion of
labeling can be render using K-means by high-scale classification gain, Equation (1).
maintaining a return index to the corresponding
each cluster.

Step 5: Afterwards, the original image is (1)

partitioned based on its k- number of color
cluster. Advanced Neural Network: There are two
significant drawbacks to the machine learning

Figure 4. Graphical Representation of Extracted Features Vector.

BIOSCIENCE BIOTECHNOLOGY RESEARCH COMMUNICATIONS An Identification and Detection Process for Leaves 1086
Farooqui & Ritika

techniques. First, they are extremely dependent of humans. ANNs are very helpful for pattern
on variables patterns as well as characteristics recognition, irrespective of any specific guidelines
to be extracted. Second, it is necessary to for recognition. Cui et al. (2018) stated that ANNs
train classifiers many times before applying to need fewer formal statistics and are capable of
applications in the real world (Zhang et.al., 2015) modeling complicated nonlinear situations. There
ANNs (Advanced Neural Networks) are the most is an increasing interest in applying ANNs to
promising tools for analyzing hyperspectral data. achieve the greater goal of disease detection in
ANNs ' mechanism is based on the nervous system wheat crop using hyperspectral data. ANNs have

Figure 5. Linear Regression between the Mean and Entropy of the

Wheat Leaf.

Figure 6. Advanced Neural Network architecture for leaf disease

detection through plain and GLCM image.

Figure 7. (A) Apply as an Input Image (B) Black Figure 8. (A) Apply as an Input Image (B) Brown
Chaff Disease Detected Rust Disease Detected

1087 An Identification and Detection Process for Leaves BIOSCIENCE BIOTECHNOLOGY RESEARCH COMMUNICATIONS
 Farooqui & Ritika

a higher degree of diagnosing ability than Neural proposed framework for the advanced neural
Networks (NNs) (Ettabaa KS, et. al.,2018). As a network represented in Fig. 6. This framework
non-invasive rapid detection method, (Zhu et helps to detect leaf diseases accurately and
al.,2017) explored the potential of hyperspectral automatically.
imaging. The Advanced Neural Networks support
the strongest ability to discriminate against plant results and discussion
diseases because they combine the finest sets
of trainers for precise classification. It can be There is various type of diseased wheat leaf
used to detect four diseases, namely black chaff, samples is taken as input. In this paper, four types
brown rust, powdery mildew, yellow rust. The of wheat leaf diseases are considered i.e. Brown

Figure 9. (A) Apply as an Input Image (B) Figure 10. (A) Apply as an Input Image (B)
Powdery Mildew Disease Detected Yellow Rust Disease Detected

Table 2. Classification results for different diseases for Proposed Method

Leaf Black Chaff Brown Rust Powdery Mildew Yellow Rust Accuracy
Disease Disease Disease Disease Disease (%)

Black Chaff Disease 30 0 0 0 100

Brown Rust Disease 0 28 0 2 93.33
Powdery Mildew Disease 0 0 30 0 100
Yellow Rust Disease 0 3 0 27 90
Average Accuracy (%) 95.83

Table 3: Comparison of different classification techniques for different

diseases. (a) Combination of the Gray-Level Co-Occurrence Matrix
and K- Nearest Neighbor Experimented values, (b) Combination of
the Gray-Level Co-Occurrence Matrix and Artificial Neural Network
Experimented values, (c) Combination of the Gray-Level Co-Occurrence
Matrix and Support Vector Machine Experimented Values.

Leaf Disease GLCM+ KNNa GLCM+ ANNb GLCM+ SVMc

Black Chaff Disease 78.02 84.00 91.11

Brown Rust Disease 80.00 85.50 90.02
Powdery Mildew 75.20 88.08 90.00
Disease
Yellow Rust Disease 78.40 85.00 91.01
Average Accuracy 77.90 85.64 90.50
(%)

BIOSCIENCE BIOTECHNOLOGY RESEARCH COMMUNICATIONS An Identification and Detection Process for Leaves 1088
Farooqui & Ritika

Figure 11. Graphical representation of Classification of different

diseases for Proposed Method.

Figure 12. Graphical representation for Comparison of different

classification techniques for various Diseases.

Figure 13. Average accuracy comparison of different

classification techniques

1089 An Identification and Detection Process for Leaves BIOSCIENCE BIOTECHNOLOGY RESEARCH COMMUNICATIONS
 Farooqui & Ritika

Rust, Black Chaff, Powdery Mildew, and Yellow systems and gathered from different resources. Our
Rust. Fig. 7 shows the input image and output experimental findings and comparisons between
image classified as wheat Black Chaff disease. distinct techniques with feature extractors
Fig. 8 shows the input image and output image showed how our sophisticated machine-learning-
classified as Brown Rust disease. Fig. 9 shows based model could effectively identify distinct
the input image and output image classified as categories of diseases in distinct crops as well
Powdery Mildew disease. Fig. 10 shows the input as provide solutions for diseases. The proposed
image and output image classified as Yellow Rust methodology was successfully tested and verified
disease. By the help of this proposed method, from the execution point of view on various types
the total of 30 sets of wheat crop leaf disease of wheat leaf diseases such as black chaff, brown
specimens is taken and categorized into four rust, powdery mildew, yellow rust. Moreover, it
broad categories of diseases which has displayed was seen that the best result could be obtained
in Table 2 and Fig. 11. Among 30 number of test resourcefully using least methodical pains to
samples of Brown Rust Diseased leaf and Yellow check the productivity of scheduled techniques.
Rust Diseased are erroneously classified as Brown Another perspective of employing this method
and Yellow Spots on the leaf, which imply 93.33% is that the plant disease can be documented
and 90.00 % of accuracy respectively. And the at the beginning or primary stage only. SVM,
other two categories of diseased leaf i.e. Black therefore, proved to be the promising method
Chaff Disease and Powdery Mildew disease are for differentiating and categorizing wheat leaf
successfully classified with 100% of accuracy. diseases with other methods by using extracted
The average accuracy of the classification of feature vector dataset. Collecting a larger number
the proposed method is 95.83%. In this paper, of samples with more variation of the diseased
classification is first done using the K-Nearest wheat leaf can extend the study. We hope that our
Neighborhood (KNN) using K-Mean’s grouping suggested scheme will contribute to agricultural
with a productivity accuracy of 77.90%. research in a provocative manner.

The finding precision is enhanced by 85.64% acknowledgements

using the advanced neural network (ANN). For
the next stage, cataloging is done through using The corresponding author wishes to thank Prof
SVM ordering process and whose productivity K.K. Raina and Prof. B. S. Panwar for the great
accuracy of 90.50%. Here also we compare with cooperation and motivation for this research.
other classification techniques with our proposed
method and examine the performance analysis References
with respect to every four types of diseases Andreas Kamilaris, Francesc X. Prenafeta-Boldu
which is illustrated in Table 3 and Fig.12. After (2018) Deep learning in agriculture: A survey,
all, among all classification techniques, the SVM Computers and Electronics in Agriculture Vol
technique result is outperforming, i.e. accuracy 147 Pages 70–90.
of 90.50%, illustrated in Fig. 13. Bannerjee, G., Sarkar, U., Das, S., Ghosh, I.,
(2018). Artificial Intelligence in Agriculture:
Conclusion A Literature Survey. International Journal
of Scientific Research in Computer Science
In this paper, we suggest a leaf disease prediction Applications and Management Studies. 7 (3),
model that can learn from the image datasets 1–6.
and then apply the different methods of machine Bashir Sabah, Sharma Navdeep. (2012) Remote
learning for disease identification of the infected area plant disease detection using image
leaf. Protection of crops in organic farming is not processing. IOSR J Electron Communication
an easy matter. Based on K-means clustering, Engg; 2(6):31–4. ISSN: 2278-2834.
SVM, advanced neural networks, advanced Cui S, Ling P, Zhu H, Keener H. (2018) Plant
machine learning models were created in our pest detection using an artificial nose system:
scheme to detect leaf diseases. Our model use a review. Sensors;18(2):378.
images that were captured by different camera Dubey, S. R., Dixit, P., Singh, N., & Gupta, J.

BIOSCIENCE BIOTECHNOLOGY RESEARCH COMMUNICATIONS An Identification and Detection Process for Leaves 1090
Farooqui & Ritika

P. (2013). Infected fruit part detection using doi.org/10.3390/sym11070939

K-means clustering segmentation technique. Rakesh Kaundal, Amar S Kapoor and Gajendra
IJIMAI, 2(2), 65-72. PS Raghava, (2006) Machine learning technique
Ettabaa KS, Ben Salem M. (2018) Adaptive in disease forecasting: a case study on rice blast
progressive band selection for dimensionality prediction,” BMC Bioinformatics.
reduction in hyperspectral images. Journal Sethy P., Negi B., & Bhoi N. (2017). Detection
of the Indian Society of Remote Sensing; of Healthy & Defected Diseased Leaf of Rice
46(2):157–67. Crop Using K-Means Clustering Technique.
Fujita, E., Kawasaki, Y., Uga, H., Kagiwada, International Journal of Computer Applications,
S., Iyatomi, H.: Basic investigation on a 157(1), 0975-8887.
robust and practical plant diagnostic system, Shuang-Qi Tian, Yong-Heng L, Zhi-Cheng C,
(2016).15th IEEE International Conference on Yong-Feng Q (2017) Effects of layering milling
Machine Learning and Applications, ICMLA, technology on distribution of green wheat main
pp. 989–992. physicochemical parameters. Journal of Food
Gebejes, A., & Huertas, R. (2013). Texture Quality Article ID 8097893: 1-7.
characterization based on grey-level co- Sanchez, P.A., and Swaminathan, M.S.
occurrence matrix. In Proceedings in Conference (2005). Cutting world hunger in half. Science
of Informatics and Management Sciences. 307,357359.doi:10.1126/science.1109057.
Jiang Lu, Jie Hu, Guannan Zhao, Fenghua Mei, UNEP (2013). Smallholders, Food Security, and
Changshui Zhang, (2017) An in-field automatic the Environment. Rome: International Fund for
wheat disease diagnosis system, Computers Agricultural Development (IFAD).
and Electronics in Agriculture Vol 142 Pages Vidyashanakara, Naveena M, G Hemnatha
369–379. Kumar (2018). Leaf Classification Based on
Konstantinos P. Ferentinos, (2018) Deep learning GLCM Texture and SVM, International Journal
models for plant disease detection and diagnosis on Future Revolution in Computer Science &
Computers and Electronics in Agriculture Vol Communication Engineering Vol 4 Issue: 3
145 Pages 311–318. Pages 156-159
Kulkarni Anand H, Ashwin Patil RK. (2012) Yitayal Abebe and Adam Bekele (2017).
Applying an image processing technique Vulnerability of Smallholder Farmers to Climate
to detect plant diseases. Int J Mod Engg Change in the Central Rift Valley of Ethiopia:
Res;2(5):3661–4. A Gender Disaggregated Approach. Ethiop. J.
Kirtan Jha, Aalap Doshi, Poojan Patel, Manan Agric. Sci. 27(2) 85-97
Shah, (2019) A comprehensive review on Zhang B, Huang W, Wang C, Gong L, Zhao C,
automation in agriculture using Liu C, et al. (2015) Computer vision recognition
artificial intelligence. Artificial Intelligence of stem and calyx in apples using near-
in Agriculture Vol 2 Pages 1–12 https://doi. infrared linear-array structured light and 3D
org/10.1016/j.aiia.2019.05.004. reconstruction. Biosyst Engg; Vol 139 Pages
LeCun, Y., Bengio, Y., Hinton, G. (2015): Deep 25–34.
learning. Nature 521(7553), 436–444. Zhu H, Chu B, Zhang C, Liu F, Jiang L, He
Marko Arsenovic, Mirjana Karanovic, Srdjan Y. (2017) Hyperspectral imaging for pre-
Sladojevic, Andras Anderla and Darko symptomatic detection of tobacco disease with
Stefanovic, (2019). Solving Current Limitations successive projections algorithm and machine
of Deep Learning Based Approaches for Plant learning Classifiers. Sci Rep;7(1):4125.
Disease Detection. Symmetry11(7),939;https://

1091 An Identification and Detection Process for Leaves BIOSCIENCE BIOTECHNOLOGY RESEARCH COMMUNICATIONS