2019 2nd International Conference on Intelligent Communication and Computational Techniques (ICCT)

Manipal University Jaipur, Sep 28-29, 2019

A New Segmentation method for Plant Disease

Diagnosis
Dr.Kiran Kumar Gurrala
Assistant Professor Lenin Yemineni∗
Department of Electronics and Communication Engineering Department of Electronics and Communication Engineering
National Institute of Technology, Andhra Pradesh National Institute of Technology, Andhra Pradesh
Tadepalligudem, 534101, INDIA Tadepalligudem, 534101, INDIA
kirankumargurrala@nitandhra.ac.in leninroyyaemineni4@gmail.com

Krupa Spandan Raj Rayana Lokesh Kumar Vajja

Department of Electronics and Communication Engineering Department of Electronics and Communication Engineering
National Institute of Technology, Andhra Pradesh National Institute of Technology, Andhra Pradesh
Tadepalligudem, 534101, INDIA Tadepalligudem, 534101, INDIA
spandanraj45@gmail.com vajjalokesh7@gmail.com

Abstract—Detecting plant diseases automatically with the help processing of image, image segmentation, extracting features
of symptoms present on leaves at earlier stage yields more and disease detection and classification.Image acquisition is
productivity in agriculture.In this paper, a novel plant disease the process of reading an image. Pre-processing of image
diagnosis method is proposed for the plants using image pro-
cessing techniques and SVM classifier. Here, disease diagnosis is done in order to enhance the quality of image which
is carried based on features extracted from the segmented will be helpful in later stages of image processing. In pre-
image after pre-processing the image of the leaves which are processing, Contrast enhancement, sharpening of image etc.,
affected with diseases.Modified color processing detection al- can be done.Image segmentation is helpful in extracting
gorithm (CPDA) is used as segmentation method to extract the features of the image.From that segmented image the
the features. SVM classifier is trained with a dataset of about
100 images of diseased leaves to identify the diseases like an- statistical parameters are extracted and that parameters are
thracnose,leafspot,leafblight,scab .For disease detection,the per- called features. These features are mean, variance, kurtosis,
formance of proposed segmentation technique is better when homogeneity etc.,Classifier classifies the disease of a plant
compared to the K-means clustering segmentation. based upon the training data given to it.
Index Terms—Plant disease, Contrast stretching, Image Seg- Arya et al proposed a method that uses Ostu segmentation
mentation, K-means clustering, Modified Colour Processing De-
tection Algorithm, SVM classifier technique along with genetic algorithm in order to detect
blight and leaf spot diseases of potato,tomato and pepper
leaves[1].Chouhan et al propped a methodology which de-
I. I NTRODUCTION
tects the diseases like common rust,leaf curl,cedar apple rust
Agriculture is responsible for providing food to mankind. It etc., using Bacterial foraging optimization based Radial Basis
is also responsible for economic development in India.So,in Function Neural Network (BRBFNN) algorithm [2].Prakash
order to get more productivity many factors like climatic et al detected the diseases of citrus leaves using K-means
conditions, water availability, plant diseases etc., play a key- clustering, GLCM features and SVM classifier[3]. Kiani et
role. Diseases that occur to plants degrade the productivity al used Colour processing detection algorithm(CPDA), Fuzzy
levels.So early detection of plant diseases and taking appro- logic classification algorithm (FLCA) to detect Strawberry
priate measures is important for proper growth of plants.Naked leaves fungal infections and iron deficiencies [4].Various
eye detection is one of method for detecting diseases of methods for segmenting image are reviewed by Kaur et al
plants which may give inaccurate results and is less effi- and have been proposed in their literature[5].Many other
cient.Also,farmers have to depend on experts who may not methods for disease detection using different techniques were
be available at the time of urgency.This method is also proposed([6],[7],[8],[9],[10],[11]).
time consuming. So,another method using image processing In this paper,image of the leaves affected with diseases like
techniques and artificial intelligence is used in order to detect blight, leaf spot, antracnose, scab of various plants like are
diseases at an early stage with more accuracy and with more detected.Contrast enhancement was done before segmenting
speed. the image.K-means clustering and proposed image segmen-
The common steps involved in modern day detection of leaf tation techniques are used to segment image of the disease
diseases using computer models are, image acquisition, pre- effected leaves.Gray Level co-occurrence matrix is used to

978-1-7281-1711-9/19/$31.00 ©2019 IEEE 137

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY ROORKEE. Downloaded on November 23,2024 at 11:57:27 UTC from IEEE Xplore. Restrictions apply.
 extract features after segementation. SVM classifier is used 1) Using K-means Algorithm: K-Means is a partitioning
to classify diseases according to the training data. method that divide a collection of objects into K groups. The
Organization of this paper in the succeeding sections is algorithm follows the following steps:
as follows.In section 2 some of the diseases that occur to • Generate K cluster centres randomly at initial.
plants are discussed.Section 3 gives the information about • Compute distance of each pixel from each kth cluster
system model that is used for disease identification in this centre
paper along with explanation of each step.Section 4 gives the • Assign each pixel to cluster it is nearest.
discussion about results of image segmentation obtained using • Modify the nearest cluster by calculating the mean of the
K-means clustering and Modified CPD algorithm.This section assigned pixel and the cluster value
also provides results - features extracted using GLCM method • Repeat these steps till the last pixel
and classification using SVM classifier.Section 5 gives the
The algorithm applied to image having intensities as I(x,y)
conclusion and future scope of the topic present in this paper.
with x = 0,1,2...M-1 and y = 0,1,2...N-1 is given as following
II. S YSTEM M ODEL steps
The system model includes various steps,from taking an (a) Initialize the K centres randomly.
image of disease effected leaf to finding the disease effected centeri = centre of the ith cluster,where i = 0,1,2...K-
by leaf of a plant. 1.
(b) Calculate the distance of pixel from each cluster centre
as
Distancei = |centrei − I(x, y)|, i = 0, 1, 2...K − 1
(1)
(c) Find the minimum distance value.
tp = minimum{Distancei }
p is the value of i at which the distance is minimum
and assigning this pixel to the corresponding cluster
centre index.
(d) Update the cluster centres
centrep + I(x, y)
centrep = (2)
2
(e) Repeat these steps till the last pixel
Fig. 1: Flow chart of the system model. After these steps, the image will be segmented into k
different images,In which each different image is having
A. Image Acquisition group of similar intensity pixel values .
It is the very first step. The plant leaf image which is to
be diagnosed is captured by digital camera and stored. The 2) Using Modified Color Processing Detection
captured image is in the RGB form. Algorithm(MCPDA): This Algorithm helps in reducing
the number of computations.In this by manipulating the pixel
B. Image Prepossessing values of image the disease effected area can be identified.
The captured image is might having the chance of low
brightness and noise, which leads to inaccurate segmentation Finally this algorithm detects red pixels of disease ef-
and classification of disease. This step includes the Image fected region in the image. By using red pixel image and
enhancement, Noise removing, Image Re-sizing etc. Image making suitable thresholds in the color processing detection
enhancement is done for increasing the contrast of image.This Algorithm, this algorithm is obtained to detect diseases more
step ensures that diseased area part will also become brighter accurately.This algorithm finds the yellow color diseases in
and sharper along with the other part of the image. the leaf of the plant at high accuracy.
C. Image Segmentation The steps involved are
This step is for detecting disease effected area in the (a) convert the given RGB image into Gray scale image
given image.That is to isolate the disease effected area from value (GSV)
whole image.There are several algorithms available for image GSV = 0.298 × R + 0.587 × G + 0.114 × B (3)
segmentation.In this paper, segmentation is done by using k-
means algorithm, Modified Color Processing Detection Algo- (b) Calculate the each R,G,B components with RIV as
rithm and a comparison of the result is made between these red intensity values of image and BIV, GIV as Blue
two segmentation intensity values of image and Green intensity values

138
Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY ROORKEE. Downloaded on November 23,2024 at 11:57:27 UTC from IEEE Xplore. Restrictions apply.
 here also the threshold should be greater than maxi-
mum repeated intensity value.
(f) Find the matched pixels between original image and
obtained binary image.which will give only diseases
effected region in the leaf image plant
The matched pixels can be found by comparing the
each component pixels of the original image and the
binary resultant image.That is, if there is a ’1’ in
a particular location of the binary image then the
intensities(each R,G,B pixel values) in that particular
location are remain same otherwise they are changed
to zero intensity values

D. Feature Extraction
From the segmented Image features will be extracted. The
features of the image are statistical parameters of segmented
image.The image is converted into Grayscale and then these
statistical parameters are calculated. The features included are
Mean, Standard Deviation, Entropy, RMS ,variance ,smooth-
ness, Krutosis, Skewness, IDM(Inverse Difference Moment),
Contrast, Co-relation, Energy and Homogeneity.All these fea-
tures are extracted from the GLCM(Gray Level Coocurrence
Fig. 2: Flow chart of the MCPD algorithm. Matrix)
P(r,s)= Probability of the pixel intensity level at the loca-
tion(r,s).The image is a gray scale image and it is having the
of image levels from 0 to 255. N = Total number of Levels = 256. μx ,
μy are the means, σx , σy are standard deviations of the image
Red component pixels(RCP) = RIV - GSV in vertical and horizontal direction. The other parameters are
Green component pixels(GCP) = GIV - GSV also calculated as
Blue component pixels(BCP) = BIV - GSV N −1 N −1
  2
 Homogeneity = P (r, s) (4)
0 IV < GSV r=0 s=0
ComponentP ixels = ,
IV - GSV else
N
 −1 N
 −1 N
 −1
2
Contrast = n2 { P (r, s) }, |r − s| = n (5)
IV is the intensity value of the image. Later modify n=0 r=0 s=0
the red pixel value as
N
 −1 N
 −1
1 2
Redpixel(RP ) = RCP − GCP
− BCP IDM = 2 P (r, s) (6)
2 2
r=0 s=0 1 + (r − s)
(c) Convert the Red pixel image into binary image with N
 −1 N
 −1
the threshold. Entropy = P (r, s) × log(P (r, s)) (7)
 r=0 s=0
0 RP < threshold
RedBinaryImage = N
 −1 N
 −1
1 else {r × s} × P (r, s) − {μx × μy }
Correlation =
r=0 s=0
σx × σ y
The threshold value should be greater than maximum (8)
repeated intensity value. That is, the threshold value is In this paper all the thirteen statistical parameters are used
mode of the intensity values as the features for the classification of the disease
(d) Find the matched pixels in the red intensity of the
image with the binary image E. Classification
By doing this the disease effected region along with Based on the Extracted Features, the disease of the plant
some other regions will be filter out from original is estimated. For the classification of the disease from the
image. extracted features a classifier is used.Here SVM(Support Vec-
(e) Again threshold red pixels of the matched image using tor Machine) Classifier is used after both the segmentation
same threshold to make the binary image. processes(K-means algorithm and modified color processing

139
Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY ROORKEE. Downloaded on November 23,2024 at 11:57:27 UTC from IEEE Xplore. Restrictions apply.
 detection algorithm).Every Classifier is having two phases (a)
Training Phase and (b) Testing Phase
(a)Training Phase:- In this Training Phase a Classifier is
Trained by giving the feature extraction from the segmented
leaf image and its corresponding disease effected by the leaf.In
SVM Classifier during this stage the weights of the vectors are
updated. (a) original (b) using k-means (c) using MCPD
(b)Testing phase:- In this phase features of the leaf which
Fig. 4: Plant Leaf and the segmented part
is to be diagnosed is given to classifier.Then it will estimate
the disease which is close to the features given and result out No of No of k-means MCPD
the name of the disease. training test im- with SVM with SVM
leaves ages classifier classifier
III. R ESULTS SCAB 100 25 16 18
BLIGHT 100 25 14 17
Different types of plant diseased leaf images are used for ANTHRACONSE 100 25 18 22
analysing the corresponding diseases effected to them by eval- LEAF SPOT 100 25 17 15
uating the K-means clustering and proposed segmentation al- TABLE I: Comparision Analysis
gorithm. K-means clustering is implemented as taking number
of clusters as three.The diseases of Scab,Blight, Anthraconse,
and Spot are used for analysis.
SVM classifier is trained using training data consisting
features of more than 100 disease affected leaves of each of
the listed diseases.Images are taken from various sources like
internet,capturing through mobile etc.With the help of features
from those images SVM classifier classifies the different
diseases accordingly. During testing stage,we provide SVM
classifier with the extracted features of the segmented image
which is to be diagnosied.Finally,classifier will display the
most suitable disease by comparing the features with the
trained data.
Here we can conclude that the proposed segmentation tech-
nique is more useful compared to k means clustering technique
for extracting features of the diseased portion,thereby it is easy
for classifier to give more accurate results. Fig. 5: Accuracy comparision

Fig 5. shows the graph which shows the comparative

analysis of disease detection using both the segmentation
techniques.

IV. C ONCLUSION
(a) original (b) using k-means (c) using MCPD
Conventional methods like naked eye detection for iden-
Fig. 3: Plant Leaf and the segmented part tifying diseases of plants fails due to slow detection and
inaccurate results. Those methods are also time consuming.In
The leaf in the Fig 3.(a) is original leaf which is effected this paper,disease diagnosis is done using image processing
by the disease Anthraconse.Fig 3.(b) and Fig 3.(c) are seg- and artificial intelligence methods.Different images of leaf
mented images using k-means clustering and Modified CPDA diseases like leafspot, scab, anthracnose, blight for many plants
segmentation respectively. were considered for identifying the disease.This paper pro-
The leaf in the Fig 4.(a) is original leaf which is effected by vides the steps involved in the process of detecting plant leaf
the disease Blight.Fig 4.(b) and Fig 4.(c) are segmented images diseases.Features are extracted after Segmentation of image
using k-means clustering and Modified CPDA segmentation using proposed algorithm. Classification of the leaf diseases
respectively. is done using SVM classifier.In this work,K-Means clustering
Table 1. shows the number of leaves that are detected and proposed algorithm for image segmentation are compared
accurately using SVM classifier after extracting features for and shown that the proposed method gives more accurate
both the segmentation techniques.For Anthraconse disease it results.This work can be extended for identification of plant
is giving more good results. leaf diseases using different hybrid segmentation techniques.

140
Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY ROORKEE. Downloaded on November 23,2024 at 11:57:27 UTC from IEEE Xplore. Restrictions apply.
 R EFERENCES
[1] Arya, M. S., K. Anjali, and DivyaUnni. ”Detection of unhealthy plant
leaves using image processing and genetic algorithm with Arduino.” In
2018 International Conference on Power, Signals, Control and Compu-
tation (EPSCICON), pp. 1-5. IEEE, 2018.
[2] Chouhan, S. S., Kaul, A., Singh, U. P., & Jain, S.” Bacterial foraging
optimization based Radial Basis Function Neural Network (BRBFNN)
for identification and classification of plant leaf
[3] Prakash, R. M., Saraswathy, G. P., Ramalakshmi, G., Mangaleswari,
K. H., &Kaviya, T. “Detection of leaf diseases and classification
using digital image processing”. In 2017 International Conference on
Innovations in Information, Embedded and Communication Systems
(ICIIECS), (pp. 1-4). IEEE.
[4] Kiani, E., &Mamedov, T.-”Identification of plant disease infection using
soft-computing: Application to modern botany”. Procedia Computer
Science, 120, 893-900 .(2017).
[5] Kaur, D., & Kaur, Y. Various image segmentation techniques: a review.
International Journal of Computer Science and Mobile Computing,
(2014). 3(5), 809-814.
[6] Singh, Vijai, and Ak K. Misra. ”Detection of plant leaf diseases
using image segmentation and soft computing techniques.” Information
processing in Agriculture 4.1 (2017): 41-49.
[7] Kiani, Ehsan, and TofikMamedov. ”Identification of plant disease in-
fection using soft-computing: Application to modern botany.” Procedia
Computer Science 120 (2017): 893-900.
[8] J. G. A. Barbedo, “A novel algorithm for semi-automatic segmentation
of plant leaf disease symptoms using digital image processing,” Tropical
Plant Pathology, vol. 41, no. 4, pp. 210 224, 2016.
[9] Khirade, Sachin D., and A. B. Patil. ”Plant disease detection using
image processing.” Computing Communication Control and Automation
(ICCUBEA), 2015 International Conference on. IEEE, 2015
[10] Lu, Y., Yi, S., Zeng, N., Liu, Y., Zhang, Y., 2017. Identification of
rice diseases using deep convolutional neural networks. Neurocomputing
267, 378–384
[11] GhaiwatSavita N, Arora Parul. Detection and classification of plant
leaf diseases using image processing techniques: a review. Int J Recent
AdvEngTechnol 2014;2(3):2347–812.ISSN (Online).

141
Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY ROORKEE. Downloaded on November 23,2024 at 11:57:27 UTC from IEEE Xplore. Restrictions apply.