Proceedings of the Third International Conference on Trends in Electronics and Informatics (ICOEI 2019)

IEEE Xplore Part Number: CFP19J32-ART; ISBN: 978-1-5386-9439-8

Computer Vision Based Turmeric Leaf Disease

Detection and Classification
A Step to Smart Agriculture
Gayatri Kuricheti Supriya P
Post Graduate Scholar Associate Professor
[email protected] [email protected]

Department of Electrical and Electronics Engineering

Amrita School of Engineering, Coimbatore
Amrita Vishwa Vidyapeetham, India

Abstract- Disease identification plays a vital role in agricultural Therefore, if the plant monitoring methods can be stored by
sector. Turmeric being a rhizomatous crop and well known for using some programming language into an automatic module
its therapeutic effects, monitoring such crops is crucial. The then the process can be error tolerant. Hence image processing
turmeric leaves are mainly exposed to diseases like Leaf Spot and plays a vital role in disease detection and analysis. In the
Leaf Blotch. The paper develops an algorithm for detecting and proposed work several image processing methods and
preventing the spreading of diseases to the whole crop and results algorithms are used for the detection of two main turmeric leaf
in high quality crop production. The data base of different leaf diseases namely Leaf Spot and Leaf Blotch and the results
images was created and processed using k-Means image obtained were classified using SVM classifiers.
segmentation and leaf images textural analysis was carried out
using GLCM. SVM classifier is used to classify the feature II. LITERATURE SURVEY
extracted images after ranking their attributes using an
information gain algorithm. A GUI has been created to portray Dr.J.Vijayakumar [1] proposed about the study of
the various stages of the image processing algorithm and detect turmeric leaf diseases and various methods to identify the
the two leaf diseases. diseases in a detailed approach. Various methods used for the
identification of diseases are Statistical analysis, Histogram
Keywords- Information Gain algorithm, GLCM, Texture Analysis, analysis and neural network based disease identification.
SVM, GUI Sagar Patil, Anjali Chandavale [2] proposed different
methods for plant disease detection. Different types of plant
I. INTRODUCTION
families are monocot and dicot family plants. In [3] it
Turmeric, a product of rhizomatous crop (Curcuma longa) mentioned about different techniques for segmentation and
which is native to tropical South Asia is belonging to the feature extraction of infected plants.
In [4] the author claims about different feature
ginger family. India is the world’s largest turmeric crop
selection techniques for the classification of turmeric plant
producer and consumes 80% of it. In the world, Indian
diseases. The RGB color image is converted into HSI color
turmeric was described as best medicinal herb because of its space and masking is applied to remove the greenness of the
implicit qualities and the presence of major bioactive leaves. Then the image is processed with K-Means
compound which is Curcumin. A city in the state of Tamil segmentation algorithm. For the feature selection different
Nadu called Erode is the most important trading area for search algorithms like principal component analysis,
turmeric, for which it is therefore called as “Yellow City”. A information gain and relief-F attribute evaluator were used.
place in Maharashtra called Sangli is the second place to erode Eapen [5], proposed a method for reducing the noise
as turmeric trading centre. in the given images and enhancing the edges before processing
it with segmentation phase. In the pre-processing stage they
In Turmeric cultivation, diseases are the most important concentrated on image resizing, histogram equalization, image
problems which reduce the harvest of Turmeric crop. The cropping and median filtering. In this approach, histogram
focused diseases in this work are Leaf Spot and Leaf Blotch. equalization was used for contrast and textural enhancement
For instant, leaf spot is affected badly in rainy seasons under of medicinal images.
humid conditions. It causes damage to a greater extent by Madhuri [6] has proposed a method to segment the
reducing rhizome size up to 52 percent. The infection of such images by masking the green pixels. The input RGB images
diseases causes the change in the color and appearance of the which are complicated to segment in three dimensional spaces,
turmeric leaves. The existing methods for plant disease therefore for the information to be available in single plane
detection simply by naked eye observation or laboratory based gray level images have been used.
techniques by experts is time consuming and requires Patil R.V. [7] says that if the number of clusters can
continuous monitoring of plant. be divided in appropriate manner, an algorithm called K-
means clustering can give accurate results. They discussed

978-1-5386-9439-8/19/$31.00 ©2019 IEEE 545

 Proceedings of the Third International Conference on Trends in Electronics and Informatics (ICOEI 2019)
IEEE Xplore Part Number: CFP19J32-ART; ISBN: 978-1-5386-9439-8

methods for edge detection for calculating the number of

clusters. The detection of edges happened using phase
congruency. Euclidean distance was used for estimating the
clusters. The popular K-means algorithm has been used for the
segmentation process. The proposed technique has been
implemented in Matlab.
In [8], they discussed about advanced computing to
assist the farmers in plant quality production. This method
used mobiles to capture diseased cotton leaves. Color feature
segmentation has been done to detect disease spots. For the
feature extraction, Edge detection was used to detect the
diseases. Neural network was implemented for classifying the
diseases.
In [9], they detected pest and disease effected leaves
by converting RGB to HSV color space and green channel has
been extracted using threshold mask. Area, Perimeter features
have been extracted. The classifier used is Decision Tree.
Fig. 3. Block Diagram of proposed Work
In [10], they differentiated between weed and a crop
which are carrot and curry leaves. The classifier used is SVM IV. METHODOLOGY
classifier.
A. Leaf Acquisition and color model Conversion
III. SYSTEM MODEL OF PROPOSED WORK
Initially, the input images are resized to the same format using
The suggested work mainly classifies the diseases in the interpolation method. The purpose of RGB to HSI conversion
turmeric leaves. The dataset of 200 turmeric leaves are is that it makes certain calculations more convenient to
collected for processing the data. The two diseases considered provide an intuitive way to identify colors. Research proves
are leaf spot and leaf blotch as shown in fig. 1 and fig. 2. that the conversions of various models speed up the image
processing with least time delays. Therefore in situations
where color description plays a vital role, HSI (Hue Saturation
and Intensity) color model is often preferred. Therefore, steps
for the conversion of color model and the output for the
conversion of leaf image into HSI are as follows

1. First read the RGB image

2. Represent it in the range [0 1]
3. where

Fig. 1. Leaf Spot Fig. 2. Leaf Blotch

The work includes different phases of image processing 4. S = 1-3R+G+B[min(R,GB,)]

techniques, the steps included are 5. I = (R + G + B)
1. Initially for pre-processing state, the input images are to be
converted from one color space which is RGB to HSI color
space and only the hue component is processed through other
phases.
2. Then K-Means Segmentation has been used for extracting
the diseased part of the leaf.
3. The output of the segmentation phase is then processed to
feature extraction phase where GLCM texture analysis is used.
Fig. 4. HSI Conversion for Diseased leaves
4. The features are trained and using multi class SVM
classifier, the diseases are classified. The system overview of B. Segmentation using k-means Clustering
the above mentioned work is as follows According to the survey, clustering methods are considered to
be very effective for color extraction. One of such clustering
techniques is K-Means [12] [13] [15]. It’s easy to implement
and requires low computational efforts. K-Means tries

978-1-5386-9439-8/19/$31.00 ©2019 IEEE 546

 Proceedings of the Third International Conference on Trends in Electronics and Informatics (ICOEI 2019)
IEEE Xplore Part Number: CFP19J32-ART; ISBN: 978-1-5386-9439-8

minimizing the sum of distances between cluster centroid and

points. The algorithm consists of following steps: 
1. Selecting K points as initial clustering centers.
2. Initiate each point of image pixels to nearest cluster As discussed in [11], various image features can be calculated
depending on its distance from cluster center. using GLCM matrix. It includes thirteen feature vectors which
3. It computes mean values for each cluster and make that are further used for disease classification. The features include
particular mean as new center. energy, contrast, correlation, entropy, and mean etc which are
4. Likewise, repeat 2 and 3 steps till the centroid no longer described below.
moves. 1. Mean

The main disadvantage of this algorithm is the need to fix the
number of clusters. For the extraction of infected part of the Mean is the basic statistical measure of an image. Mean
leaf, 3 clusters (representing background, healthy part and the computes the mean of the image array as rows and as
infected part) can be ideal. The disease attacked on the green columns and is denoted as µ and it tends to remove the
pigments of leaf changes the color at that portion. The noise of the image.
segmentation results are shown where Fig.5 is the input 2. Standard Deviation
image; Fig.6-8 is the clustered image where one of the clusters It is the statistical index which quantifies mean and also
will be the required disease portion, after removing all the portrays the diversity in the image. It is defined as:
unwanted part. This Region of Interest (ROI) is then
transmitted to the next phase for applying feature extraction

algorithms.
3. Entropy
Entropy= 

Entropy shows the amount of information of an image. It

measures the randomness of an image texture. When the
image is texturally uniform, entropy is large.

4. RMS
RMS Value represents the image distribution of an
image and it is given by:


Fig. 5. Input Image
5. Variance

Variance= 

Where µ is the mean

Variance is the second order statistical feature which
measures the heterogeneity. Variance will increase when
there is difference in the grey level values from the mean
value.

Fig. 6. Cluster1 Fig. 7. Cluster2 Fig. 8. Cluster3

6. Smoothness
It defines the relative smoothness of intensity of an image
and is carried out as replacing every pixel by the weighted
A. Feature Extraction and Dataset formation average of its neighborhood.
Texture analysis is a method to define the relationship among
the pixels in local area. Gray Level Co-occurrence Matrix is 7. Kurtosis
the texture analysis statistical approach which considers the K= 
pixels spatial relationship. By finding the pairs of pixel with
certain values obtained in specific orientation and distance, 8. Skewness
GLCM characterizes the textural properties of an image.
GLCM is a way to calculate the statistical features of a given S= 
image. It can tell specific properties regarding the spatial
Where p (z) is the corresponding histogram of the
closeness of gray-levels in an image. This method is best used
for classifying texture features. For a given image M*M, the grey levels of an image with i= 0, 1…………L-1
elements of GLCM matrix ‘M’ are defined as

978-1-5386-9439-8/19/$31.00 ©2019 IEEE 547

 Proceedings of the Third International Conference on Trends in Electronics and Informatics (ICOEI 2019)
IEEE Xplore Part Number: CFP19J32-ART; ISBN: 978-1-5386-9439-8

9. Inverse Difference Movement kernel functions in which linear kernel has been implemented
IDM=  to gain the accurate classification.

V. RESULTS
10. Contrast
Contrast=    For result visualization, a MATLAB GUI (Graphical User
Interface) has been created that shows the original input image
11. Correlation getting converted to HSI color space and the enhancement
takes place. Hue and Saturation values will be used and
Correlation=   
segmented using K-Means Clustering in which K=3 are made
appropriate. For the feature extraction phase, GLCM
Where is the standard deviation techniques have been used for the extraction of statistical
features with necessary formulae. The data set has been
Correlation measures the linear dependency of the
created for different leaf diseases and classified using
neighbor pixels of an image with gray levels.
supervised learning algorithm called Support Vector Machine
(SVM). The GUI outputs showing diseased leaf and healthy
12. Energy
leaf classification are shown in the fig. 9 and fig. 10.
Energy=   

Where M is the GLCM matrix; i and j are the rows and

columns of the matrix
Energy explains the disorders in textures. It is the sum of
the squares of the pixels in the obtained GLCM matrix
where M is the matrix. When the pixel repetition is more,
energy value is high. Energy can reach a maximum value
of one. It is also called angular secondary moment.

13. Homogeneity is defined as

  

Such 13 features can be calculated from the GLCM matrix

that describes texture of an image as described by Haralick.
The feature vector dataset is created for all the disease classes.
Among the 13 attributes few features can be overlapped
making classification difficult. Therefore, feature selection
algorithm is used where it minimizes the number of features
that are considered for disease classification. Fig. 9. Leaf Spot Classification
Feature selection is an algorithm that enables automatic search
for best attributes from the trained data. Trained data may
contain many features that are irrelevant to the classification
task; therefore Information Gain algorithm is used for the
ranking of the features. Based on the ranking method, it gives
the subset of suitable attributes for the classification. The
classifier used for the proposed work is SVM Classifier.

B. Support Vector Machine Classifier

Based on the survey [14], different classifiers such as Back
Propagation Neural Network (BPN), Artificial Neural
Network (ANN), Fuzzy Logic and Support Vector Machine
(SVM) which shows 91% accuracy rate has been chosen for
the classification of diseased leaves.
SVM is a supervised learning model that analyzes the data for
classification analysis. In the set of trained data, where each
example belongs to one or the other group of diseases, such
algorithm develops a model that makes new data fit into one
or the other category. As with any supervised algorithm, SVM
will be trained and then cross validated. Then, the new data
will be classified using the trained machine. There are various Fig. 10. Healthy Leaf Classification

978-1-5386-9439-8/19/$31.00 ©2019 IEEE 548

 Proceedings of the Third International Conference on Trends in Electronics and Informatics (ICOEI 2019)
IEEE Xplore Part Number: CFP19J32-ART; ISBN: 978-1-5386-9439-8

VI. CONCLUSIONS [15] W. Cui and Y. Zhang, "Graph based multispectral high resolution image
segmentation," in Proc. International Conference on Multimedia Technology
The paper was modeled for the detection and classification of (ICMT), pp. 1-5, 2010.
turmeric leaves. It involved various processing techniques and
classification algorithms for the accurate results. The data set
has been created with 200 leaf images using .mat file. SVM
was implemented for classification of the turmeric leaf
diseases and its accuracy has been tested and found to be high.
In future, better automatic algorithms can be implemented
using IOT platform with large number of data set.

REFERENCES

[1] Dr.J.Vijayakumar, “Study of Turmeric Plant Diseases and Methods of

Disease Identification using Digital Image Processing Techniques,”
International Journal on Future Revolution in Computer Science and
Communication Engineering, vol 3, 2017.

[2] Sagar Patil, and Anjali Chandavale, “A Survey on Methods of Plant

Disease Detection,” International Journal of Science and Research, vol 4,
2015.

[3] Darshana A, Jharna Majumdar, and Shilpa Ankalaki, “Segmentation

Method for Automatic Leaf Disease Detection,” International Journal of
Innovative Research in Computer and Communication Engineering, vol 3,
2015.

[4] Pream Sudha V, “Feature Selection Techniques for the Classification of

Leaf Diseases in Turmeric,” International Journal of Computer and
Technology, vol 43, 2017.

[5] Eapen M, and Korah R, “Medical image segmentation for Anatomical

Knowledge Extraction,” Journal of Computer Science, vol 10, 2014.

[6] Jadhav Sanjivani, Lohar Utkarsha, Bhagat Madhuri and Salunke

Shubhangi, “Plant Disease Identification System for Android,” International
Journal of Computer Science and Mobile Computing, vol 5, 2016.

[7] R. Patil and K. Jondhale, "Edge based technique to estimate number of

clusters in k-means color image segmentation,” in Proc. 3rd IEEE
International Conference on Computer Science and Information Technology
(ICCSIT), pp. 117-121, 2010.

[8] P. Revathi and M. Hemalatha, “Advance Computing Enrichment

Evaluation of Cotton Leaf Spot Disease Detection Using Image Edge
detection,” IEEE, 2012.

[9] Aparna K and Supriya P, “Precision Agriculture in Maize Fields,”

International Conference on Intelligent Computing and Control Systems,
2018.

[10] Rani, Anjali K, Supriya P and T V Sarath, “Computer Vision based

Segregation of Carrot and Curry leaf plants with weed identification in Carrot
Field,” International Conference on Computing Methodologies and
Communication, pp. 185-188, 2017.

[11] R. M. Haralick, K. Shanmugam and I. Dinstein, "Textural Features for

Image Classification," in IEEE Transactions on Systems, Man, and
Cybernetics, vol. SMC-3, no. 6, pp. 610-621, Nov. 1973.

[12] Puri C, and Sukhwinder Singh, “Image segmentation and Classification-

A Review,” International Journal of Innovative Research in Science,
Engineering and Technology, vol 3, (apr) 2014.

[13] F. C. Monteiro and A. Campilho, "Watershed framework to region-based

image segmentation,” in Proc. International Conference on Pattern
Recognition, ICPR 19th, pp. 1-4, 2008.

[14] M. Hameed, M. Sharif, M. Raza, S. W. Haider, and M. Iqbal,

"Framework for the comparison of classifiers for medical image segmentation
with transform and moment based features," Research Journal of Recent
Sciences, vol. 2277, p. 2502, 2012.

978-1-5386-9439-8/19/$31.00 ©2019 IEEE 549