IRJET- Leaf Disease Detecting using CNN Technique

INTERNATIONAL RESEARCH JOURNAL OF ENGINEERING AND TECHNOLOGY (IRJET) E-ISSN: 2395-0056
VOLUME: 06 ISSUE: 05 | MAY 2019 WWW.IRJET.NET P-ISSN: 2395-0072
© 2018, IRJET | Impact Factor value: 7.211 | ISO 9001:2008 Certified Journal | Page 6953
LEAF DISEASE DETECTING USING CNN TECHNIQUE
**Prof Ramya C N, *Naveen G C, *Aiswariya Dev S, *Sucharitha N N,
*Department of Electronics and Communication Engineering, Atria Institute of Technology, Bengaluru,
India
**Asst Professor, Department of Electronics and Communication Engineering, Atria Institute of
Technology, Bengaluru, India
---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract--Identification of plant disease is very difficult in
agriculture field. If identification is incorrect then there is a
huge loss on the production of crop and economical value
of market. Leaf disease detection requires huge amount of
work, knowledge in the plant diseases, and require the
more processing time. Therefore, we can use image
processing for identification of leaf disease in MAT LAB.
Identification of disease follows the steps like loading the
image, contrast enhancement, converting RGB to HSI,
extracting of features and SVM. We have proposed system
has used to design and implementation Digital image
processing techniques for detecting, quantifying and
classifying plant diseases using SVM and KNN with ANN
Algorithm the K-means clustering technique for the
segmentation purpose and Artificial Convolution Neural
Network (CNN) technique for the classification of the
mango, pomegranate ,guava, sapota leaf disease. The
system as been tested with the different numbers of test
data set collected from different regions. This system has
tested for different numbers of clusters to get the optimal
number of cluster that can produce the best performance of
the proposed leaf disease identification and control
prediction system. This proposed system has overcome the
problem of identification of mango leaf disease manually.
Key words---K-Means Clustering, ANN Convolution
method, Leaf Disease
1. INTRODUCTION
Digital image process is the use of computer algorithms
to perform image process on digital pictures. It permits a far
wider vary of algorithms to be applied to the computer file
and might avoid issues like the build-up of noise and signal
distortion throughout process. Digital image process has
terribly important role in agriculture field. It is widely
accustomed observe the crop disease with high accuracy.
Detection and recognition of diseases in plants
mistreatment digital image method is extremely effective in
providing symptoms of characteristic diseases at its early
stages. Plant pathologists will analyze the digital pictures
mistreatment digital image process for diagnosing of crop
diseases. Computer Systems area unit developed for
agricultural applications, like detection of leaf diseases,
fruits diseases etc. altogether these techniques, digital
pictures are collected employing a camera and image
process techniques are applied on these pictures to extract
valuable data that are essential for analysis. The diseases
are mostly on leaves and on stem of plant. They are
Potassium, Magnesium, Calcium, Zinc, or iron deficiencies
due to insects, rust, nematodes etc. on plant. It is important
task for farmers to find out these deficiencies as early as
possible. Following example shows that how deficiencies on
plant Leafs reduces the productivity from Image processing
techniques is been used to detect on mango, pomegranate,
guava, sapota etc.
2. LITERATURE REVIEW
Describing the identification of various leaf diseases as
illustrated and discussed below. [1] An identification of
variety of leaf diseases using various data mining
techniques is the potential research area. The diseases of
different plant species has mentioned. Classification is done
for few of the disease names in this system. The concept
SVM for classification is used in this system. This work finds
out the computer systems which analyzed the input images
using the RGB pixel counting values features used and
identify disease wise and next using homogenization
techniques, Sobel and Canny using edge detection to
identify the affected parts of the leaf spot to recognize the
diseases boundary is white lighting and then result is
recognition of the diseases as output. [2] In this proposed
system, grape leaf image with complex background is taken
as input. Thresholding is deployed to mask green pixels and
image is processed to remove noise using anisotropic
diffusion. Then grape leaf disease segmentation is done
using K-means clustering. The diseased portion from
segmented images is identified. Best results were observed
when Feed forward Back Propagation Neural Network was
trained for classification. [3] The feature extraction is done
in RGB, HSV, YIQ, and Dithered Images. The feature
extraction from RGB image is added in the suggested
system. A new automatic method for disease symptom
segmentation in digital photographs of plant leaves. The
diseases of different plant species has mentioned.
Classification is done for few of the disease names in this
system. The disease recognition for the leaf image is
performed in this work. An identification of variety of leaf
diseases using various data mining techniques is the
potential research area. The diseases of different plant
species has mentioned. Classification is done for few of the
disease names in this system. The concept SVM for
classification is used in this system [4] In this section the
recent trends in using CNN and deep learning architectures
in agricultural application are discussed. Prior to the advent
of deep learning, image processing and machine learning
techniques have been used to classify different plant
diseases (Barbedo 2013; Pydi- pati, Burks, and Lee 2005;
Camargo and Smith 2009b; 2009a). Image processing
techniques, such as image enhancement, segmentation,
color space conversion, and altering, are applied to make

the images suitable for the next steps. Then important
features are extracted from the image and used as an input
for the classier (Al-Hiary et al. 2011). The overall
classification accuracy is therefore dependent on the type of
image processing and feature extraction techniques used.
However, latest studies have shown that state of the art
performance can be achieved with networks trained using
generic data. CNNs are multi-layer supervised networks
which can learn features automatically from datasets. For
the last few years, CNN’s have achieved state-of-the-art
performance in almost all important classification tasks. It
can perform both feature extraction and classification under
the same architecture (Atabay 2016b). [5] Xu et al. (2011)
proposed a method to detect nitrogen and potassium
deficiencies in tomato plants. The algorithm begins
extracting a number of features from the color image. The
color features are all based on the b* component of the
L*a*b* color space. The texture features are extracted using
three different methods: difference operators, Fourier
transform and Wavelet packet decomposition.[6] in this
paper S.jeyalakshmi and R. Radha Explains Plants and crops
require 13 essential mineral nutrients to grow and survive.
They acquire these nutrients from the soil. Deficiency of
these nutrients affects the growth and quality of the
plant/crop. Thus, diagnosing nutrient status of minerals
plays a crucial role in agriculture and farming. Nutrient
deficiency symptoms in plants/crops would normally be
visible in leaves. These symptoms include interveinal
chlorosis, marginal chlorosis, uniform chlorosis, necrosis,
distorted edges, reduction in size of the leaf etc[7]
Sjadojevic et al. have presented the concept of deep
convolution neural network (CNN) and fine tuning for the
identification of plant leaf diseases. Authors have
considered thirteen different types of dataset images with
healthy leaf images for the experimentation. Deep learning
based Caffe framework as been used along with the set of
weights learned on a very large dataset by authors. The core
of framework developed in C++ and provides command line,
Python, and MAT LAB interfaces. Authors have used the 10-
fold cross validation test for the accuracy assessment. The
overall results shows the accuracy of 96 % and precision
value lie between 91 % to 96 %.Fine-tuning has not shown
significant changes in the overall accuracy, but
augmentation process had greater influence to achieve
respectable results.[8] Mohanty et al. have used the concept
of deep convolutional neural network for the analysis of
plant leaf diseases. Authors have used the Plant Village
dataset having 38 classes based 54, 306 images for the
experimentation. This dataset consists of 14 crop species
and 26 types of disease-affected plants. Deep learning based
architecture of Alex Net and Google Net have been
considered. Training mechanism of transfer learning and
training from scratch approaches had been using. Testing
has also performed with different ration aspect of training
and testing. Authors have shown the accuracy of 99.35 %
for the disease analysis. However, there were some
limitations with the concept. Approach is limited to applied
dataset and presented approach is not able to detect the leaf
diseases if the leaf side changed apart from the front area.
3. PROBLEM IDENTIFICATION
By a detail study of literature, we have identified the
following problems:
The leaf disease as identified manually. In this technique,
a man who has the information of the plant leaf as been
called for examination for the diseased plant then the leaf
diseases will be distinguish by the learning and that
individual advises experience of that individual and the
control. This all procedure happens physically so the time it
now, prolonged and has a great deal of shots of being
misguided judgment of right leaf diseases identification
proof.
Until now, the various automated systems have been
developed for the cotton, grape, banana, bamboo, rice, herb
leaf diseases but there is not any system that can
automatically detect leaf diseases, so this proposed work
will provide an automated system, which will be able to
identify leaf deficiencies and predict the appropriate control
for the leaf disease.
4. PROPOSED SYSTEM AND EXPLANATION
The proposed method for the identification and control
prediction of mango, pomegranate, guava, sapota leaf
deficiencies is composed of image processing using
convolution neural network techniques. The framework of
proposed approach are made for detection of deficiencies,
two image databases are required, one for training purpose
and other for testing. In addition, for deficiencies detection,
Image preprocessing is required for enhancing images. The
next step is image segmentation is required; otherwise, the
feature of non-infected region will dominate over the
feature of infected region. After segmentation, feature
extraction done from segmented image and finally the
training and classification are performed. Each step of
proposed system is discussed in this section The block
diagram of proposed mango, pomegranate, guava, sapota
leaf disease identification and control prediction algorithm
is shown in block diagram
Block diagram of proposed leaf disease detecting

This method utilizes the techniques of image processing
and neural network in a composite manner to obtain the
desired goal. The proposed leaf deficiencies identification
and control prediction algorithm consist of the following
steps:
Step 1: Image Acquisition
The camera is vertically oriented and approximately a
distance of 0.5-meter distance to be maintained while
capturing the images. Image quality is definitive for the after
effects of investigation, influencing both the ability to detect
features under examination and accuracy of consequent
estimations. Image enhancement techniques are used to
emphasize features of interest and highlight certain details
hidden in the image. To improve the quality of the image,
preprocessing steps are applied over the image. MAT LAB
version is used for implementation of the digital image
processing algorithms. Mango, pomegranate, guava, sapota
leaf images are captured from different regions by using
digital mobile camera, are used for training and testing the
system then the background data are removed and stored in
standard jpg format.
Step 2: Image Pre-Processing
Preprocessing of the image includes shade correction,
removing artifacts, and formatting. Some images, originally
from camera, manifest uneven lighting called shade. Due to
variation in outdoor lightning conditions, some regions are
brighter and some others are darker than the mean value
for the whole image. This phenomenon is a consequence of
inaccuracy in the system. Precise tuning of camera is done
to minimize this effect. The images contain some artifacts
induced like scratches, coat, or mark, lumps of dust or
abrasive particles. Hence, median filter and infielder is been
used to remove such artifacts. Formatting deals with
storage representation and setting the attributes of the
image. The images acquired from the camera are of 1920 x
1080 pixels and reduced to suitable size for the reasons of
reducing computational time required for feature extraction
and their storage on the medium. Image pre-processing
includes the following three modules:
 Cropping leaf image.
 Resize.
 Median filter.
Step 3: Image Conversion
The image conversion includes the following types of
conversion for different purposes:
 RGB to gray.
 Gray to binary.
 RGB to L*a*b* color shape.
Step 4: Segmentation
Image segmentation used to serrate the distinct parts
with some information in the image. K means clustering
method used for the proposed method.
K Means Segmentation
K-Means clustering algorithm classifies the input data
points into many number of classes based on clusters
inherent distances. The algorithm assigns that data features
to create a vector space for clustering. These data points
have clustered around centroids.
∑ ∑ ( )
Where k is number of clusters Si, I = 1, 2,…k and µi is the
mean or centroids of all points
Algorithm Steps:
1. Computing the histogram based on the intensities.
2. Initialize the centroids with k random intensities.
3. Perform the steps until the cluster labels of the
images reaches constant.
4. Clustering is done based on distance from the
intensities of centroids to the cluster intensities
from the c
| |
5. New centroids of each cluster is computed
∑
∑
Where k is the number of clusters to be found, I number
of iterations, k-means clustering is performed to split the
image into three clusters. In these three clusters, one or two
clusters resemble the diseases, which will give the
segmentation. To extract the ROI in diseased mango leaf the
K-means clustering algorithm is used. This algorithm
clusters the point nearest to the centroids. The centroids is
basically the average of all the points in that cluster and has
coordinate as the arithmetic mean over all points in the
cluster, separately for each dimension.
Step 5: Feature Extraction
A pattern can denote a quantitative or morphological
description of an object or some other point of interest in an
image, in which some organization of underlying structure
can be supposed to live. In other words, a pattern is an
arrangement of descriptors. Descriptors are also called
features in pattern recognition literature. Only significant
features are extracted from the processed image. This is
where the features reduction method is adopted. In the
present work, feature extraction employs color features
based on RGB, HSI color models, texture features based on
GLCM
The following features are extracted to classify the
disease:

1) Area: The actual number of pixels in the region of
interest.
2) Orientation: The angle θ (in degrees ranging from -90
to 90 degrees) between the x-axis and the major axis of the
ellipse that has the same second- moments as the region
√
3) EquivDiameter: It specifies the diameter of a circle
with the same area as the region. Computed as:
4) Extent: It specifies the ratio of pixels in the region to
pixels in the total bounding box. Computed as:
rea of OI
(3)
5) Solidity: It specifies the proportion of the pixels in the
convex hull that are also in the region and computed as:
Solidity=
rea
(4)
6) Convex Area: It specifies the number of pixels in
'Convex Image'.
7) Major Axis Length: It specifies the length (in pixels) of
the major axis of the ellipse that has the same normalized
second central moments as the region.
8) Number of Objects: It is the number of white pixels,
which are disconnected to each other in binary image. Color
feature extraction
Color feature extraction
One of the primary facets of color feature extraction is
the selection of a color space. A color space is a
multidimensional space, in which different dimensions
represent different constituents of the color. Most color
spaces are three-dimensional. An instance of a color space is
RGB, which attributes to each pixel a three element vector,
giving the color intensities of the three primary colors,
namely, red (R), green (G), and blue (B). The space covered
by the R, G, and B values completely describes visible colors,
which are entitled as vectors in the 3D RGB color space.
Therefore, the RGB color space offers a useful starting point
for representing color features of the images the following
method is adopted in the extraction of RGB features. The
foremost step is the separation of RGB components from the
original color images. The next step is the computation of
mean, standard deviation, variance, and range from the
separated RGB components using the following Equations
∑
Where,
N is the total number of panels,
Xi is the ith pixel value
⁄ ∑ √
Maximum element and minimum elements from given
input color (RGB) image is calculated using Equation (3).
max1=max (image), max2=max (max1) (4)
The above function returns the row vector containing
maximum element from each column, similarly find
minimum element from whole matrix using Equation (2) to
(4)
min1=min (image),min2=min(min1) (5)
Range is the difference between the maximum and
minimum elements and is given in the Equation (2) to (6).
Range=max2-min2 (6)
When humans see a color object, the object is depicted by its
hue (H), saturation (S), and brightness or intensity (I). Hue
is a good descriptor of a pure color (pure yellow, orange or
red), whereas saturation refers to the amount of pure color
mixed with white light. The chromatic notion of intensity
(gray level) which describes brightness is the most useful
descriptor of monochromatic images. The intensity
component is easily quantifiable and interpretable. The HSI
color model separates the intensity component from the
color carrying information (hue and saturation) in a color
image. Therefore, the HSI model is an absolute aid for
developing image processing algorithms based on color
descriptions that are natural and instinctive to humans,
who, after all, are the developers and users of these
algorithms. The hue, saturation, and intensity components
are extracted from the RGB components RGB color space
can be transformed to HSI color space using the Equations
(7) to (10).
Color feature reduction
It is found through experimentation that only eight color
features, which are common in all the sample images, are
significant. Hence, these eight features contribute more to
the classification of plant diseases. Therefore, eight features
have been considered as first-level feature reduction. The
reduction is done based on threshold and delta value. Any
feature values below the threshold are discarded. The
threshold is chosen based on average of minimum feature
value and maximum feature value. The threshold value is
empirically determined as 0.2. Delta is the minimum
difference between two feature values and is empirically
determined as 10-3 .The procedure involved in color feature
reduction is given in the Algorithm 1
Algorithm1: Color feature reduction
Input: color (RGB) image.
Output: Reduced color feature vector. Se
Description: Delta is the minimum difference between two
features and is set to 10-3 Threshold is the average of
minimum and maximum feature value and is set to 0.2
Start

Step 1: Separate the RGB components from the original 24-
bit input color image
Step 2: obtain the HIS components using the Equation (7),
(8),(9) and (10)
Step 3: Compute mean, variance, and range for each RGB
and HIS components using the Equation (1) through (6)
Step 4: Threshold= (minimum feature value +maximum
feature value)/2
Step 5: Initialize feature vector to zeros
Step 6: For(i = 1 to size of the feature vector) if (value of
feature(i)>threshold) Select as reduced feature
Step 7: For (i= 1 to size of the reduced feature vector)
Compare each feature with the other if (feature values are
equal OR feature values differ by data) Discard the feature
Else Select as reduced color feature Stop.
Texture feature extraction
For texture features based on spatial domain analysis,
one way to describe the descriptor is using a second order
statistics of pairs of intensity values of pixels in an image
using co-occurrence matrix method [89]. The co-occurrence
matrix method of texture description is developed using
spatial gray level dependence matrices (SGDMS), which is
based on repeated occurrence of some gray level
configuration in the texture. This configuration varies
rapidly with distance in fine texture and slowly with coarse
textures. The GLCM Pφ, d (i, j) represents a matrix of
relative frequencies describing how frequency pair of gray
levels (i, j) appear in the window separated by a given
distance d= (dx, dy) at an angle ‘φ’ [105]. Gray level co-
occurrence matrices (GLCMs) method counts how often
pairs of gray level of pixels separated by certain distance
and oriented in a certain direction, while scanning the
image from left-to-right and top-to- bottom. In the present
work, a distance of 1 (d=1) when ‘φ’ is 0° or Equations (11)
to (16) are used to evaluate the textural features
∑ ∑ (11)
∑ (12)
∑ ∑ (13)
Maximum Probability =max (P(x, y)) (14)
∑ ∑ (15)
The differentiation between sample images is carried out in
the simplest way, quantifying average gray levels within the
matrix, change in the gray level with respect to average level
of minimum and maximum gray levels present in the
matrix. Hence, basic co-occurrence features, namely, mean,
variance, and range has been considered using the
Equations (1) to (6).
Texture feature reduction
It is found through experimentation that only five
texture features, which are common in all the sample
images, are significant. Hence, these five features contribute
more to the classification of plant diseases. Therefore, five
features have been considered as first-level feature
reduction. The reduction is done based on threshold and
delta value. Any feature values below threshold are
discarded. The threshold is chosen based on average of
minimum feature value and maximum feature value. The
threshold value is empirically determined as 100. Delta is
the minimum difference between two feature values and is
empirically determined as 10-3 [8].The procedure involved
in texture feature reduction is given in the Algorithm 2
Algorithm 2: Texture feature reduction
Input: Color (RGB) image.
Output: Reduced texture feature vector
Description: Pφ, d (x, y) means GLCM matrices in the
direction (φ=00, 450, 900, and 1350) and‘d’ is the distance.
Delta is the minimum difference between two features and
is set to 10-3. Threshold is the average of minimum and
maximum feature value and is set to 100.
Start
Step 1: For all the separated RGB components, derive the co-
occurrence matrices Pφ, d (i, j) in four directions 00, 450,
900, and 1350 and d=1
Step 2: Compute mean, variance, and range for each RGB
components using the Equations (1) through (6)
Step 3: Threshold = (minimum feature value + maximum
feature value)/2
Step 4: Initialize feature vector to zeros
Step 5: For (i =1 to size of the feature vector) If (value of
feature (i) >threshold) Select as reduced feature
Step 6: For (i=1 to size of the reduced feature vector)
Compare each feature with the other If (feature values are
equal OR feature values differ by delta)
Discard the feature
Else
Select as reduced texture feature
Stop.
Step 6: Classification
The symptoms of plant disease exhibit different
properties like color, shape, and texture. When samples of
different normal and disease affected
agriculture/horticulture crops are considered, patterns vary
from disease to disease. Color is an important dimension of
human visual perception that allows discrimination and
recognition of visual information. Many natural surfaces and
naturally occurring patterns reveal texture characteristic,
meant to capture the granularity and repetitive forms of
surfaces within an image that considered work has used
some state of the art color and texture features for
recognition and classification of diseases affected
agriculture/horticulture crops to validate the accuracy and
efficiency. For the classification, the CNN Neural Network
classifier technique is used which consist of three layers
namely input layer, a hidden layer, and an output layer. The

study has adopted artificial neural network based classifiers
using CNN classifiers in the recognition of images of plant
disease and studied their behavior in terms of suitability of
classifiers for identification of different plant diseases.
Step 7: Disease Identification and Control Prediction the
CNN assigns an appropriate mango leaf disease class i.e.
Potassium, magnesium, calcium, and zinc or iron leaf spot.
Then it appropriate control prediction for the bacterial leaf
spot or red rust gives by the system automatically. The
process of recognition and classification is given in the
Algorithm 3.
Percentage = correctly recognized sample images
Accuracy (%) total number of test sample images x100
Algorithm 3: Recognition and classification of plant diseases
affecting agriculture/horticulture crops
Input: Colour (RGB) images of plant diseases affecting
agriculture/ horticulture crops.
Output: Recognized and classified images.
Start
Step 1: apply color, texture feature extraction input color
image, obtain color, and texture features
Step 2: apply color and texture feature reduction Algorithms
1 and 2 to color, texture features, and obtain reduced color
and texture feature vector
Step 3: Train the SVM and CNN with reduced color and
texture feature vector
Step 4: Accept test images and repeat Steps 1 and 2
Step 5: Recognize and classify the images using SVM and
CNN Stop.
II. SYSTEM REQUIREMENT SPECIFICATIONS
1. Operating System: Window
2. Software: MAT LAB
3. Programming language: Embedded C
III. HARDWARE REQUIREMENTS SPECIFICATIONS
1. Main processor: Intel i7 Core
2. Hard Disk Capacity: 1 TB
3. Cache memory: 500 MB
5. EXPERIMENTAL RESULTS
The experimental environment is worked on a 2.23 GHz
Intel(R) Core(TM) i7 CPU M730 with 4 GB of RAM PC. By
using computer simulation, “M TL B we are performaning
the leaf deficiencies identification
SVM based Pixel classifier
A support vector machine (SVM) is used to recognize
plant disease affecting agriculture/horticulture crops. The
study has chosen SVM because of its efficient
implementations and performances proved to be excellent
for high dimensional problems and small data sets. Viewing
training input vector in an n-dimensional space, SVM
constructs a hyper-plane in the space, which can be used for
classification that has the highest distance to the closest
training data point of any class (functional margin). To
compute the margin, two parallel hyper- planes are
constructed, one on every side of the isolating hyper-plane,
which are pushed up in opposition to the two data sets. The
aim is to determine which class a new data point belongs
based on data points associated to one of the two classes. In
the case of support vector machines, a data point is
computed as a p-dimensional vector (a list of p numbers)
and it is meant to know whether such levels can be forked
by a (p−1) dimensional hyper-plane. This is called a linear
classifier or maximum margin.
Figure 1.1 shows the different Hyperplane
K-NN Deficiency Classifier
These test images are pre-processed by using median
filter and the output pre-processed now these pre-
processed images are converted into the binary images
based on the threshold value. These binary images are now
used for the segmentation in which the K-means clustering
method is used; here the number of cluster taken is 3 and
the clusters formed by K-means clustering method The
classifiers are trained and tested using images of plant
diseases. The sample images are divided into two halves
and one half is used for training and other is used for
testing. The colour and texture features are used to train
and test neural network model. The percentage accuracy of
recognition and classification is defined as the ratio of
correctly recognized sample images to the total number of
sample images. Value of K can be chosen in runtime based
on the diseased leaf image.

Fig. K-Means segmentation of RBG image
K-means clustering is simple and computationally faster
than other clustering techniques and it works for large
number of variables. However, it produces different cluster
result for different number of number of cluster and
different initial centroids values. So it is required to
initialize the proper number of number of cluster k and
proper initial centroids
Fig 1.2 K-Means segmentation of hue image
CNN based classifier
The study has considered CNN as a model to identify
plant disease symptoms affecting agriculture/horticulture
crops. Leaf diseases image database is created by acquiring
images under challenging conditions such as illumination,
size, pose and orientation, using a Mobile camera of
resolution 4608 x 3456. It consists of 1200 images of both
diseased and healthy leaves. The diseases include
Potassium, Magnesium leaf spot, leaf gall, leaf Webber, leaf
burn of plant. In order to reduce the computational time
complexity, the images are resized from the size 4608 x
3456 to 256 x 256. The proposed CNN architecture consists
of an image input layer followed by three hidden layers and
then the output layer. The layer implementation is
represented in Table 1.
TABLE I. LAYER IMPLEMENTATION OF THE CNN MODEL
Figure 2.4. Leaf imagecropping and resize example.
(a) Input image,
(b) Cropping image, (c) 229 × 229 image.
The leaf images of size 256 x 256 x 3 are given as input
to the input layer. Data augmentation is performed in order
to increase the dataset by generating artificial data. The
images are then passed through the hidden layers. Each
hidden layer consists of a convolutional layer, batch
normalization layer, Rectified Linear Unit followed by the
max pooling layer. Feature extraction is performed using
convolutional and pooling layers, whereas classification is
performed by the fully connected layer. Each convolutional
layer and pooling layer consists of different number of
filters, of varying size. The three convolution layers consists
of 32, 64, 128 filters of size 11x11, 7x7, 5x5 respectively
with stride 2 and padding. The batch normalization layer
and the ReLU layer increase the training process and
network performance. The three max pooling layers
consists of 5x5, 3x3 and 3x3 filters respectively with stride 1
and padding, P=1 for maxpooling layer 1 and P=0 for
maxpooling layers 2 and 3. Then 50% dropout is employed
to deactivate the least learned features. The features learnt
by the convolutional and pooling layers are then classified
by using two fully connected layers of size 64 and 6
respectively. The size of the second fully connected layer is
equal to the number of classes. It specifies the probability
distribution for each class. Steepest Gra dient Descent
algorithm is used to train the proposed
CONCLUSION
. The proposed CNN based leaves disease identification
model is capable of classifying four different deficiencies in
leaves from the healthy one. Since CNN does not require any
tedious preprocessing of input images and hand designed
features, faster convergence rate and good training
performance, it is preferred for many applications rather
than the conventional algorithms. The classification
accuracy can be further increased by providing more images
in the dataset and tuning the parameters of the CNN model.
Result
Figure 1.3 & 1.4 shows the graphical representation
means through graph explain about the CNN & SVM system
as showing accuracy of leaf diseases in percentage and
grading point. CNN is also a classifier, which is used for
testing the training datasets like neural network. There is
difference only in their approaching ways and how the data
does is selected and defined.

REFERENCES
[1] Youssef Es-saady, Ismail El Massi, Mostafa El Yassa,
Driss Mammass and Abdeslam Benazoun, “ utomatic
recognition of plant leaves diseases based on serial
combination of two SVM classifiers” 2nd International
Conference on Electrical and Information
Technologies(ICEIT), IEEE, 2016.
[2] Sanjeev S. Sannakki, Vijay S Rajpurohit, V. B. Nargund
and PallaviKulkarni, “Diagnosis and Classification of
Grape Leaf Diseases using Neural Networks”,
International Conference on Computing
Communications and Networking Technologies IEEE,
2013.
[3] Lumb, Manisha, and Poonam Sethi, Texture Feature
Extraction of RGB, HSV, YIQ and Dithered Images using
GLCM, Wavelet Decomposition Techniques,
International Journal of Computer Applications, 68
(11), 2013
[4] Atabay, H. A. 2016b. A convolutional neural network
with a new architecture applied on leaf classiﬁcation.
IIOAB J 7(5):226–331.
[5] Xu G, Zhang F, Shah SG, Ye Y, Mao H. Use of leaf color
images to identify nitrogen and potassium deficient
tomatoes. Pattern Recognit Lett. 2011;32(11):1584–
1590. doi: 10.1016/j.patrec.2011.04.020.
[6] A review on diagnosis of nutrient deficiency symptoms
in plant leaf image using image processing by
S.jeylakshmi and R. radha ICTACT journal on image and
video processing,May 2017, volume:07, issue:04 ISSN:
0976-9102,DOI: 1021917/ijivp.2017.0216.
[7] Sladojevic, Srdjan, Marko Arsenovic, AndrasAnderla,
DubravkoCulibrk, and DarkoStefanovic. "Deep Neural
Networks Based Recognition of Plant Diseases by Leaf
Image Classification." Computational Intelligence and
Neuroscience 2016 (2016).
[8] Mohanty, Sharada P., David P. Hughes, and Marcel
Salathé. "Using Deep Learning for Image-Based Plant
Disease Detection." Frontiers in Plant Science 7 (2016)
[9] Leaves Classification Using SVM and Neural Network
for Disease Detection by Bhushan R. Adsule, Jaya M.
Bhattad vol 3, issue 6, june 2015 ISSN:2320-9801,DOI:
10.15680.
[10] Davoud Ashourloo, Hossein Aghighi, Ali Akbar Matkan,
Mohammad eza Mobasheri and mir Moeini ad, “ n
Investigation Into Machine Learning Regression
Techniques for the Leaf Rust Disease Detection Using
Hyperspectral Measurement” IEEE Journal Of Selected
Topics In Applied Earth Observations And Remote
Sensing, pp.1-7, May 26, 2016.
[11] Youssef Es-saady, Ismail El Massi, Mostafa El Yassa,
Driss Mammass and bdeslam Benazoun, “ utomatic
recognition of plant leaves diseases based on serial
combination of two SVM classifiers” 2nd International
Conference on Electrical and Information
Technologies(ICEIT), IEEE, 2016.
[12] Sonali Dash, K.Chiranjeevi, Dr.U.R.Jena and
akula.Trinadh, “Comparative study of image texture
classification technique”,International Conference on
Electrical, Electronics, Signals, Communication and
Optimization IEEE, 2015.
[13] Barbedo, G.C.A. (2013) Digital image processing
techniques for detecting quantifying and classifying
plant diseases, Springer Plus, 2:660.
[14] Carmago, A. and Smith, J.S. (2009) Image pattern
classification for the identification of disease causing
agents in plants, Computers and Electronics in
Agriculture, 66(2009), p. 121-125.
[15] Chaerle, L., Lenk, S., Hagenbeek, D., Buschmann, C., Van
Der Straeten, D. (2007) Multicolor fluorescence
imaging for early detection of the hypersensitive
reaction to tobacco mosaic virus, Journal of Plant
Physiology, 164(3), p. 253-262.
[16] Kulkarni, A. and Patil, A. (2012) Applying image
processing technique to detect plant diseases,
International Journal of Modern Engineering Research,
2(5), p. 3361-3364.
[17] Lopez, M.M., Bertolini, E., Olmos, A., Caruso, P., Gorris,
M.T., Llop, P., Penyalver, R., Cambra, M. (2003)
Innovative tools for detection of plant pathogenic
viruses and bacteria, International Microbiology, 6, p.
233-243.
[18] Purcell, D.E., O’ Shea, M.G., Johnson, . ., Kokot, S.
(2009) Near infrared spectroscopy for the prediction of
disease rating for Fiji leaf gall in sugarcane clones,
Applied Spectroscopy, 63(4), p. 450-457.
[19] Sankaran, S., Mishra, A., Eshani, R. and Davis, C. (2010)
A review of advanced techniques for detecting plant
diseases. Computers and Electronics in Agriculture, 72.
[20] Schaad, N.W. and Frederick, R.D. (2002) Real time PCR
and its application for rapid plant disease diagnostics,
Canadian Journal of Plant Pathology, 24(3), p.250-258.
[21] Spathis, C., Georgakopoulou, K., Petrellis, N. and Birbas,
A. (2014) Integrated microelectronic capacitive
readout subsystem for lab-on-a-chip applications, IOP
Measurement Science and Technology, 25, 055702.
[22] Arivazhagan, S., R. NewlinShebiah, S. Ananthi, and S.
Vishnu Varthini. "Detection of unhealthy region of
plant leaves and classification of plant leaf diseases
using texture features." Agricultural Engineering
International: CIGR Journal 15, no. 1 (2013): 211-217.
[23] Kranz, J. "Measuring plant disease." In Experimental
techniques in plant disease epidemiology, ISSN no.
978-3-642-95534-1, page no. 35-50. Springer Berlin
Heidelberg, 1988.
[24] James, W. Clive. "Assessment of plant diseases and
losses." Annual Review of Phytopathology Vol. 12, issue
no. 1, page no. 27-48, 1974.
[25] Khirade, Sachin D., and A. B. Patil. "Plant Disease
Detection Using Image Processing." In Computing
Communication Control and Automation (ICCUBEA),
2015 International Conference on, pp. 768-771. IEE

IRJET- Leaf Disease Detecting using CNN Technique

More Related Content

What's hot (20)

Similar to IRJET- Leaf Disease Detecting using CNN Technique (20)

More from IRJET Journal (20)

Recently uploaded (20)

IRJET- Leaf Disease Detecting using CNN Technique