Early Detection of Crop Diseases

Prof. Mitali Shewale, Suraj Bhosale, Ganesh Chavan

Mitali.shewale@somaiya.edu, suraj.vb@somaiya.edu, ganesh.chavan@somaiya.edu

Information Technology Department, KJ Somaiya College of Science and Commerce, Mumbai, Maharashtra, India

Abstract 1. INTRODUCTION
Crops square measure being affected by uneven Modern technologies have given human society
atmospheric condition resulting in faded agricultural the flexibility to supply enough food to satisfy the
yield. This affects international agricultural economy. demand of quite seven billion folks. However, food
Moreover, condition becomes even worst once the security remains vulnerable by many factors
crops square measure infected by any malady. together with temperature change, the decline in
Agriculture not solely provides food for the human pollinators, plant, and others. Plant diseases aren't
existence; it's conjointly an enormous supply for the solely a threat to food security at the world scale
economy of any country. several bucks square however may also have unfortunate consequences
measure being spent to safeguard the crops annually. for farmer farmers whose livelihoods rely upon
One technique to shield the crop is early pesterer healthy crops. within the developing world, quite
detection in order that the crop will be protected eighty % of the agricultural production is generated
against pesterer attack. If pests square measure by farmer farmers and reports of yield loss of quite
detected, applicable measures will be taken to shield five hundredth because of pests and diseases
the crop from an enormous production loss at the top. square measure common. The hindrance and
Early detection would be useful for minimizing the management of disease have perpetually been
usage of the pesticides and would offer steerage for wide mentioned as a result of plants square
the choice of the pesticides. ancient technique of measure exposed to outer surroundings and
examination of the fields is oculus examination square measure extremely susceptible to diseases.
however it's terribly troublesome to own a close Normally, the correct and speedy designation of
examination in massive fields. to look at the total malady plays a very important role in dominant
field, several human specialists square measure disease, since helpful protection measures square
required that is extremely costly and time measure typically enforced once correct
overwhelming. Hence, associate automatic system is designation.
needed which may not solely examine the crops to
The agricultural land mass is quite simply being a
observe pesterer infestation however can also classify
feeding sourcing in today’s world. Indian economy
the kind of pests on crops. The planned system
is extremely dependent of agricultural
determines region of interest from associate input
productivity. Therefore, in field of agriculture,
image. The input image should undergo following
detection of malady in plants plays a very
stages: Image Acquisition, Image pre-processing,
important role. To observe a disease in terribly
Image segmentation so as to get region of interest.
initial stage, use of automatic malady detection
Thus, the region of interest was with success
technique is useful.
determined.
The existing technique for disease detection is
Keywords: Image Acquisition, Image pre-processing,
solely oculus observation by specialists through
Image Segmentation, K-means rule
that identification and detection of plant diseases

1
is completed. For doing therefore, an oversized with the quantity of pesticide/insecticide square
team of specialists similarly as continuous measure sent to the farmer and also the farmer will
observation of plant is needed, that prices terribly see the main points on the web site. this might
high after we do with massive farms. At a similar prove edges in observation massive fields of crops,
time, in some countries, farmers don't have correct and therefore mechanically observe the symptoms
facilities or perhaps concept that they'll contact to of diseases as shortly as they seem on plant leaves.
specialists. because of that consulting specialists
even price high similarly as time overwhelming too. 2. LITERATURE REVIEW
In such conditions, the steered technique proves to M Dhakate of Reference [1] projected a pomegranate
be helpful in observation massive fields of crops. disease detection system using image processing and
Automatic detection of the diseases by simply neural network. Numerous pomegranate diseases are
seeing the symptoms on the plant leaves makes it mentioned together with bacterial Blight, Fruit Spot,
easier similarly as cheaper. disease identification Fruit rot and Leaf spot. Grey level co-occurrence matrix
by visual approach is a lot of grueling task and at a (GLCM) is developed to get statistical texture options.
similar time, less correct and might be done solely
in restricted areas. Whereas if automatic detection Precision Agriculture has been employed in Reference
technique is employed it'll take less efforts, less [2] for pomegranate plant farming. Author has
time and become a lot of correct. mentioned regarding best usage of water in irrigation,
correct nutrient management to plant and avoid crop
Image process may be a technique to perform losses because of diseases and pests with correct
some operations on a picture, so as to urge planning of sprays.
associate increased image or to extract some
Reference [3] presents a summary of existing
helpful info from it. Advent of recent technologies
reportable techniques helpful in detection of diseases
like Digital image process and Image analysis
of agricultural merchandise. A comparative study of
technology has several applications within the
various strategies based on the type of agricultural
biological field. regarding seventy eight of the
product, methodology and its potency together with
farmers square measure little and marginal within
the benefits and downsides is additionally mentioned.
the country and that they square measure poor in
resources. Therefore, {they square Hidden Markov Model is introduced in Reference [4]
measure|they're} not during a position to use for developing a observance system which is able to
optimum amount of inputs in their crops that are determine diseases on grapes crop in its early stages.
essential for increasing the productivity. Most of Numerous environmental conditions like temperature,
farmers might not apprehend the quantity of humidity, leaf condition amount, etc. are examined by
chemical needed for crops and therefore it's going Hidden Markov Model to predict the disease on Grape
to result in unbalanced use of chemical and that plant.
they can also not apprehend that
Anderson Rocha at Reference [5] introduced a new
pesticide/insecticide to be used for the unhealthy
system for multi-class fruits/vegetables classification in
crop. thus the yield gets affected.
super market scale. A total of 2633 images of 15
A easy net application might to some extent categories of produce were collected in the 5 months
facilitate farmer solve the matter of a farmer to period. Image processing techniques were applied in
observe a malady. The farmer clicks the image of order to extract bag-of-features which consisted
the crop and sends it. The image is processed statistical color, texture and structural descriptors. The
mistreatment the Image process techniques and system was developed in order to categorize fruits and
also the malady is detected. the main points of the vegetables based on the extracted features. The
malady and also the space affected in conjunction proposed method combined many features and

2
classifiers. The results showed a reduction of the tested for 82 different biotic and a biotic stress using a
classification error up to 15% with respect to baseline. large database containing images of symptoms
belonging to, affecting the leaves of 12 different plant
Reference [6] detects, diagnose and prescribe control
species.
options, an expert’s system using Rule- Based
Algorithm to identify Plant Diseases in the Philippines. Reference [12] detect little leaf disease is a hazardous
This application helps farmers to identify problem in disease found in pine trees. The author found,
rice plants and perform action to detect and diagnose detection of plant disease through some automatic
rice plant disease and prescribe feasible control technique is beneficial and it reduces a large work of
options. monitoring in big farms of crops. Image segmentation
and soft computing techniques is used to detect the
Reference [7] detects disease in tomato leaf, using
plant disease.
automatic technique to minimize the effect of
presence of vein, RGB image should be colour
transformed before segmentation. After then Otsu
threshold can be applied on colour component to 3. METHODOLOGY
detect disease spot accurately. 3.1 System Block Diagram:
To classify the crop species Reference [8] develop a Block diagram showing implemented modules to find
model for deep learning for image-based plant disease Region of Interest (Infected Area) from given input
detection using deep convolution neural network with image of leaf or fruit:
images of plant leaves with the goal to identify disease
on images.
Reference [9] classifies crop disease based on image
processing, developed grading method based on
computer image processing technique, firstly it
acquires the images of the crop disease, leaves. Vector
median filtering method is used to pre-process the crop
leaves. Using statistical pattern recognition method for
segmentation and then it calculates the ratio
percentage of the lesion area and the leaf area. Finally
crop disease harm degree and classification standard
were determined.
Reference [10] has conferred upon unhealthy region of
plant leaves. Initially colour transformation is
performed for RGB image as input, and green pixels are
masked, remove specific threshold value followed by
the segmentation process. Then texture statistics are
computed for the useful segments, after that texture
features are to classify plant leaf diseases. Finally, the
extracted features are passed through the classifier. 3.1.1 Image Acquisition:

Reference [11] has identified multiple plant diseases In Acquisition Process, Images of the infected areas of
using digital image processing; suggest colour crops are capture through the camera by user. This
transformation and colour histograms method for image is in RGB (Red, Green and Blue) form.
disease identification, after that pair wise-based
classification system is used. Its performance was

3
3.1.2 Image Pre-processing: Enhanced image is then sent to next module for
further processing.
In image pre-processing step, removal of noise and
distortion from image takes place. Non-local means
denoising for preserving textures algorithm of scikit-
image library is used to remove noise from image.
Image enhancement is carried out for increasing the
Contrast.

Original Pre-processed
3.1.3 Image Segmentation:
Segmentation means partition of image into diverse 3.2.2 Image Segmentation:
part of same skin tone or having some likeness
The Preprocessed image from previous module is now
dissection means parcelling of picture into different
segmented to separate regions of same likeness. Here,
part of same elements or having some likeness. Here,
the k-means clustering algorithm is used to segment
the segmentation is carried out using k- Means
image into k clusters. The value of k is taken as 4,
Clustering algorithm. Then Region of Interest is
hence total 4 different clusters are obtained after
extracted from segmented image.
segmentation of which one cluster contains majority
of diseased part (Region of Interest).
3.1.4 Feature Extraction:
Feature extraction plays an important role for
identification of an object. In many application of
image processing feature extraction is used. Color,
texture, morphology, edges etc. are the features which
can be used in plant disease detection. In this paper we
are considering color, CCV and morphology as a feature
for disease detection. They have found that Pre-processed Segmented
morphological result gives better result than the other
features.. 3.2.3 Separate Clusters from Segmented Image:

3.2 Module wise Implementation:

3.2.1 Image Pre-Processing:

The input image to the system may contain some Cluster 0 Cluster 1
noise and distortions. So before processing further, it
is necessary to pre-process image. In this module,
input image is pre-processed by removing noise and
distortions from image (using Non-local means
denoising for preserving textures algorithm) which
results into image enhancement.

Cluster 2 Cluster 3

4
Cluster 3 contains majority of diseased part, hence it is account as constant and τ’s price is assumed as 1
the Region of Interest Chronicles of image. Any Connected element has range
of pixels over or capable capable its pixels area unit
3.2.4 Feature Extraction thought-about coherent and also the others area unit
incoherent.
In this approach we've used color, morphology and
K-Means Image Segmentation Algorithm
color coherence vector for feature extraction

• A color is wide used visual feature and principally Step 1. Read input image.
wont to compare pictures. A color bar chart represents
the distribution of colourize image. whereas computing Step 2. Transform image from RGB to L*a*b*
the pixels of various colours in a picture, if the colour color space.
area is giant, then the colour area is split into sure
Step 3. Classify colors using K-Means
range of tiny intervals. every tiny interval is named bin.
clustering in 'a*b*' space.
Then by numeration range of pixels in every of the bins,
we tend to get color bar chart of image. Here, we tend Step 4. Label each pixel in the image from the
to computed color bar chart for all pictures in dataset results of K-Means.
and save in information which can be used for
comparison of question image with dataset pictures. Step 5. Generate images that segment the image
Usually, color bar chart of 2 pictures is compared by color.
exploitation total of square of variations. Here 3 bin bar
chart is constructed for RGB color area. Step 6. Select disease containing segment.

• Morphology is tool used for extracting image parts. In this experiment, squared Euclidean distance is used
These extracted image parts area unit helpful in for the K-means clustering. We use L*a*b* color space
description and illustration of region form like because the color information in the L*a*b* color
boundary extraction. Here, the erosion idea is space is stored in only two channels (i.e. a* and b*
employed that is prime operation of morphology for components), and it causes reduced processing time
getting the boundaries of pictures. when applying for the defect segmentation. In this experiment input
erosion operation, we tend to get image boundary by images are partitioned into four segments. From the
subtracting worn pictures from original image. By empirical observations it is found that using 3 or 4
cluster yields good segmentation results.
exploitation morphology, we'll extract illness form
vector from healthy fruit.
4. RESULT AND ANALYSIS
• Color Coherence Vector CCV may be a histogram-
based methodology for comparison pictures that For our experiment a database of several diseased and
comes with spatial info. during this technique, every healthy images created. The RGB image is pre-
constituent during a given color bucket is assessed as processed to remove noise and distortions from image
either coherent constituent or incoherent constituent. and then K- means clustering segments the image into
Classification of every constituent relies on whether or 4 clusters. Here, k=4 is considered because it gives
not or not it's a part of an oversized similarly-colored proper clusters than k=3 or 5. One of the clusters from
region. Coherent pixels area unit a part of some sizable obtained clusters contained majority of diseased part
contiguous region, whereas incoherent pixels don't which is our Region of Interest. This Region of Interest
will further undergo feature extraction and
seem to be belongs to some sizable region. so as to
classification to detect disease present in crop. Here,
reason CCV, 1st image is blurred. Then color area is
result is Region of Interest from original image of
discretized, guaranteeing that there area unit solely n infected fruit or leaf.
distinct colours within the image. Here, τ is taken into

5
 detection and classification of agricultural products,"
2016 International Conference on Computing
Communication Control and automation (ICCUBEA),
Pune,2016,pp.16.doi:10.1109/ICCUBEA.2016.786003

[4] S. S. Patil and S. A. Thorat, "Early detection of

grapes diseases using machine learning and IoT," 2016
Second International Conference on Cognitive
Computing and Information Processing (CCIP), Mysore,
Original Region of Interest 2016, pp.1-5. doi: 10.1109/CCIP.2016.7802887.

[5] Rocha, Anderson, et al. "Automatic fruit and

5.CONCLUSION AND FUTURE vegetable classification from images." Computers and
SCOPE Electronics in Agriculture 70.1 (2010): 96-104.

Thus, Region of Interest (Infected Area) is successfully [6] Roselia C. Morco, Fredilyn, B. Calanda,
extracted from input image of fruit or leaf. These Jonathan A. Bonilla, Mark Jade S. Corpuz, Junnel
Region of Interest will further undergo feature
extraction. For more accuracy wavelet features will be E. Avestro, Jean M. Angeles (2017) e-RICE: An Expert
extracted from these regions using GLCM (Grey Level System using Rule-Based Algorithm to Detect,
Co-occurrence Matrix). Extracted features then Diagnose, and Prescribe Control Options for Rice Plant
classified with dataset features to detect disease Diseases in the Philippines 978- 1-4503-5392,2017
present in crop. Then respective treatment will be
suggested for found disease with approximate amount [7] S. Budihal, Sandhya R, Soumya D Hajawagol,
and cost of pesticides needed. (2015), Detection of Disease in Tomato Leaf,4pp.2278-
5140.
These early stage detections of disease and providing
appropriate treatment will result into increase in yield [8] Sharada Prasanna Mohanty, David Hughes,
of crop. Marcel Salathe (2016) Using Deep Learning for Image-
Based Plant Disease Detection,2016.

REFERENCES [9] Youwen Tian, Lide Wang, Qiuying Zhou (2012)

Grading Method of Crop Disease Based on Image
[1] M. Dhakate and Ingole A. B., "Diagnosis of Processing, pp. 427–433, 2012.
pomegranate plant diseases using neural network,"
2015 Fifth National Conference on Computer Vision, [10] S. Arivazhagan, R. Newlin Shebiah, S. Ananthi,
Pattern Recognition, Image Processing and Graphics S. Vishnu Varthini (2013) Detection of unhealthy region
(NCVPRIPG), Patna, 2015, pp. 1-4.doi: 10.1109/NCV of plant leaves and classification of plant leaf diseases
PRIPG.2015.7490056 using texture features,15pp11-23,2013

[2] S. More and M. Nighot, "An agro advisory for [11] Jayme Garcia Arnal Barbedo, Luciano Vieira
pomegranate field using wireless sensor network," Koenigkan, Thiago Teixeira Santos (2016), Identifying
2016 International Conference on Automatic Control multiple plant diseases using digital image
and Dynamic Optimization Techniques (ICACDOT), processing,147pp104-116
Pune,2016,pp.557562.doi:10.1109/ICACDOT.2016.787
7647 [12] V. Singh, A.K. Misra (2017) Detection of plant
leaf diseases using image segmentation and soft
[3] M. K. Tripathi and D. D. Maktedar, "Recent computing techniques,4PP41-49,2017
machine learning based approaches for disease

6
Plagiarism Report:
This Plagiarism report is generated using the website duplychecker.com. The content in this paper is 98 %
unique.

7
8