A Review on Disease Detection in Plant, Leaf and

Fruit Using Image Processing Technique

Dheresh Soni1
[email protected], Assistant Prof., School of CSE, VIT University, Bhopal, India
Richard Essah2
[email protected], Research Scholar, Department of CSE, Chandigarh University, Chandigarh, India
Jayram Dwivedi3
[email protected], Assistant Prof., Department of CSE, SIRT College, Bhopal, India
Mathias Agbeko4
[email protected], Research Scholar., ICT Education, University of Education, Winneba-Ghana
Gurinder singh5
[email protected], Assistant Prof. CSE, Chandigarh University, Mohali, India

Abstract - The many image processing techniques for There are many techniques for early detection of plant
spotting leaf disease are discussed in this research diseases. Traditionally, plant disease identification is done
paper. Plant leaf diseases may be promptly detected by hand, which is inefficient for large harvests. Using
and categorized using digital image processing. The digital image processing to diagnose plant diseases saves
research combines clustering and artificial neural time and money. Insecticides and time are saved. Various
networks (ANNs) to categorize disorders (used by papers propose digital image processing for reliable plant
several authors). We are currently looking on novel disease diagnosis. Several scholars have developed image
image processing methods and ways to recognize leaf processing techniques [1]. This research paper describes
illness. This research paper also covers the financial many types of Image processing is used to detect and
and agricultural losses brought on by fruit disease. categorize leaf and fruit diseases. Experts used to detect
Previously, ill fruit could only be found manually, but and identify illnesses. Professionals advise farmers on how
technological advancements have made picture to control the disease. To save time and money in
processing possible. Training and testing are the two developing countries, automated disease screening
stages. All information relating to infected and techniques were developed. Agriculture uses technology to
uninfected fruit is retained during the training phase in boost production. Fruit yield and quality using this
order to determine whether the fruit is ill and, if so, the technique may help farmers avoid financial losses.
illness. The methods used to identify contaminated fruit
are described in this research paper. These methods Automatic fruit disease detection is critical since it detects
support early disease detection for fruit farms. symptoms early on, resulting in less economic loss per
fruit. By monitoring current fruit conditions, producers
Key Words: CCV, LBP, K-Means Clustering, Neural may take preventive measures for next year. Some diseases
Network, Fuzzy Logic, SVM may damage the tree's shoots and branches [2].
Agrobacterium, rhinocladium, anthracnose powdery
1. INTRODUCTION mildew, and red rust are all prevalent mango diseases.
Diseases caused by pectobacterium and agrobacterium.
India is an agricultural nation, and agriculture employs the Rhinocladium produces mango nausea. Anthracnose is
majority of the people. Farmers have a broad variety of caused by humidity, temperature fluctuations, and heavy
Fruit and Vegetable crops to choose from. Technological rain. Powdery mildew blooms when rain or mist coupled
assistance may help to enhance farming. Pathogens induce with a chilly night. Alga causes red rust. Because mangoes
disease in plants under any environmental situation. In are irregular in shape, eye inspection is inadequate.
most instances, illnesses are visible on the plant's leaves,
fruits, and stems; therefore, disease identification is critical 2. LITERATURE REVIEW
to crop cultivation success. Pathogens, microorganisms,
fungus, bacteria, viruses, and other microorganisms cause Leaf Blast (MagnaportheGrisea) and Brown Spot
the majority of plant illnesses. Plant diseases are (MagnaportheGrisea) are the subject of the research paper
sometimes caused by an unfavorable environment, which [3]. (CochiobolusMiyabeanus). To extract traits from
includes soil and water. diseased leaf regions, the author first takes a photograph. It
uses a SOM (Self Organising Map) neural network to
zoom in on pictures of sick rice. Create a SOM input vector enhancement. Color pictures are enhanced using discrete
in two ways. It employs no padding and interpolates cosine transforms. These convert color spaces, but in
missing points. Fifty-fifths zooming uses interpolation to feature extraction, the GLCM and graycoprops functions
normalize spot size It doesn't help to alter the frequency of are used by the author to compute statistics like contrast,
the The zooming algorithm correctly classifies test images. energy, homogeneity, and entropy.
With image processing, they can detect leaf and stem
illness [4]. The author used five plants from the Jordanian Citrus leaf diseases are classified using SVMRBF and
al-Ghor leaf collection. Using image processing, you can SVMPOLY. Processing images to identify [11] proposes
see early burns and mold, as well as late burns and mold, as an research paper on orchid leaf disease. Leaf diseases in
well as very little white spots. This technique captures orchids include solar sunburn and black leaf mark. First,
images and then uses K-Means clustering to segment them. collect pictures and save them on a computer. The pictures
Then CCM is utilized to obtain textural characteristics are pre-processed by adjusting histograms, intensities, and
from diseased leaf and stem. filtering Boundary segmentation removes tiny things while
retaining big ones [12]. Segmentation uses a line's start and
Plant diseases are classified via neural network end points to trace edges. The ROI was added to the GUI.
propagation. Image processing shows that plant diseases The image is then categorized using white pixels. This
may be properly detected and classified (about 93 percent). method produces accurate results with little error. Research
Paper [5] utilized LABVIEW and MATLAB to identify paper [13] suggested image processing to detect illnesses in
chili plant disease. This method detects sick leaves. Stage 1 tomato leaves. Image capture Photograph ill tomato leaves,
IMAQ LABVIEW recorded the image. MATLAB and with early blight and powdery mildew. These include
Vision perform more image processing. For example, edge reducing noise, resizing, isolating and eliminating
detection and Fourier filtering are both examples of picture background. The author used Gabor wavelet and SVM to
preprocessing. Color clustering is used in feature extraction identify and classify tomato diseases. The Gabor wavelet
to tell chili leaves apart from non-chili leaves [6]. Then the transform creates feature vectors for categorization.
health of each chili plant is determined using image
recognition and categorization. Less hazardous chemicals To detect tomato illnesses during classification, an SVM is
in the chili manufacturing means cheaper costs and better trained. SVM creates decision trees using feature vectors
quality chili. and classes. Then came tomato leaf blight. These kernels
are used in SVM. On-the-fly grid search with N-fold cross-
Image processing for detecting Malus Domestica in leaf validation Pre-processing removes noise and other
disease histogram is used. Histograms calculate grayscale elements from pictures. This is done by converting RGB
picture intensity [7]. In image segmentation, Co-occurrence images to greyscale using the equation f(x) = 0.2989*R +
matrix technique examines texture, whereas K-means 0.5870*G + 0.114*B. Images are segmented for sick leaf
clustering analyzes color. Texture analysis examines a identification utilizing boundary and spot detection. K-
picture's texture. The color gap between objects and the means clustering classifies items [14]. From greyscale
centroid of their class or matched cluster is reduced via pictures, Otsu threshold generates binary images. We can
color analysis. A pixel's value is compared to a threshold to tell a lot about plant diseases just by looking at them.
identify an item. Photo comparisons of texture and color Extract of HB leaf Color co-occurrence and components
detect plant diseases. Bayes and K-means clustering are are used in image processing [12]. An ANN or back
planned [8] and propose image processing for plant propagation network classifies illnesses [15].
bacterial infection detection. Bacterial leaf scorch is an
early-stage plant infection. Image capture and conversion
to a computer-readable format are examples of picture
acquisition. Then K-means clustering separates the
foreground and background images.

The leaf area is emphasized by eliminating the base

pictures' leaf clusters. K-means clustering, in contrast to
fuzzy logic [9], is straightforward, quick, and low-
maintenance. FPGA tools can use ADSP target boards.
With image processing, they can detect damaged citrus leaf
regions [10]. These illnesses include canker and wilt.
Anthracnose symptoms include citrus greening and
Figure1: Architectures of Common Artificial Neural
overwatering. To start a database, the author suggested
taking high-quality photos using a digital camera. Pre-
processing does color space conversion and image
Network (Single layer and multi-layer architectures) [15] resolution. Morphological methods may eliminate noise.
Image processing is proposed in Research paper for the Prior to removing shadows and adjusting images, shadow
detection of Scorch and Spot plant diseases. First, an RGB reduction is essential for segmentation and feature
image of the plant is taken. To convert RGB color data to extraction.
space, preprocessing is required, and then green-pixel
masking is performed. Inside infected cluster borders, Step 2 - Image Segmentation: Identifying the area of
unmask cells. Segmenting an image to remove useful parts interest using specific algorithms, or generating clusters of
By extracting color, texture, and edge characteristics. areas by comparing adjacent pixels.
Neuronal networks classify illnesses. Research in the field
will evaluate disease status of citrus plants [16]. Paper [15] Thresholding: When the backdrop and foreground contrast
suggests the use of image processing to detect illness in strongly, this approach is employed. Within a global
groundnut plants. Cercospora leaf spots appear in the early threshold, all pixels are marked ‘0' and regarded in the
and late stages of the season (Cercosporidiumpersonatum). background, while pixels marked ‘1' are considered in the
foreground. The Otsu thresholding technique uses the
HSV colorization is used to leaf RGB images. Finding infraclass variance pixels to calculate the threshold value.
green pixels in a picture saves time. Co occurrence
matrices for color and texture extraction Texture feature Region Growing: Using the area growth technique, a seed
extraction analyzes texture pictures in two ways. The first is chosen based on the surrounding pixels' properties. This
is structured, the second is statistical. This work used method's seed selection is critical. R.O.I. expands based on
statistics. Back propagation classifies infectious groundnut pixel similarity.
illnesses. Back propagation has two steps: 1) a wide range
of weights They were 97% accurate in their diagnosis of
four different diseases. Then, for the RGB leaf image,
create a color transformation structure and map RGB color
values to the structure's spatial coordinates. The picture is
then split using K-means [17]. The green pixel masking
step eliminates the unneeded space. To remove the leaf's
green section, At this point, the infected cluster's borders
are masked cells that have been calculated. With each
cluster, an HSI matrix is generated.

The GLCM function determines feature and texture

statistics in step 4. The characteristics are then fed into a
pre-trained neural network [17]. They propose using image
processing to identify sugarcane leaf disease [14]. They are
looking at Red Rot, Downy Mildew and Sugarcane
Mosaic. Image acquisition captures pictures in TIF, PNG,
JPEG, and BMP formats. RGB pre-processing is done in
the pictures, it then converted in grayscaled where
unnecessary data is eliminated. Image segmentation
identifies healthy (green pixels) and contaminated (red
pixels) regions. The feature extraction methods include
linear SVM, nonlinear SVM, and multiclass SVM.

3. System Design
The picture should go through the following processes in
order to identify disease:

1) Pre-Processing.
2) Segmentation-Process. Figure 2. Schematic representation of the SVM algorithm
3) Feature Extraction-Process. classification process.
4) Classification-Process.
k-mean Clustering: k is the number of clusters in this
Step 1 - Image Pre-processing: This phase removes noise technique. It's based on real-world data. Each observation
and improves image quality. The super resolution is an object to K-mean. K-mean clustering identifies a
technique may make a low-resolution image high- partition in which elements in one cluster are closer to one
another than in other groups. the gap between each item Using image processing techniques may help identify leaf
and its clustered centroid is reduced iteratively. diseases early on. An artificial neural network (ANN) is a
kind of machine learning algorithm. This research paper's
Fuzzy C-mean Clustering: Iterative algorithms are methods save time and provide results.
employed in c-mean clustering. Initially, fuzzy cluster
centers are calculated, and a fuzzy partition matrix is Fruit Detection System's proposed technique based on
created. The goal function is decreased in the following fuzzification was tested, and it rapidly identified the feature
iteration to determine the best location for the cluster. for pattern recognition system using fuzzy curves and
When the maximum number of iterations is reached, the fuzzy surfaces Characteristics that are reliant on other
iteration process is automatically terminated. important features are eliminated on fuzzy surfaces. It had
proven efficient in the extraction of features. With the K-
Step 3 - Feature extraction: It reduces the number of mean clustering technique, segmentation may be done
sources needed to create a dataset by extracting useful quickly and accurately. Because it accurately maps input
information from the input picture. data, SVM has proved to be the best classification method.
Dimensions feature space-linear or nonlinear mapping
Color: Color is a stable visual characteristic. The color techniques. For low-quality images, the Intent Search
systems utilized are RGB and HIS. A fruit's infection Technique has been shown to be the most effective way to
may be recognized by its color. Color image processing improve the image.
has three basic principles.
REFERENCES
Areas: The alteration of color Individual color plane
spatial processing, Color vector processing 1. J. G. ArnalBarbedo, “Digital image processing
techniques for detecting, quantifying and classifying
Texture: Texture in a photograph describes how colors plant diseases,” SpringerPlus, vol. 2, article 660, pp.
are distributed in an image. Fruits with diseases have a 1–12, 2013.
different feel. So we can use texture to assess illness 2. A. J. Tatem, D. J. Rogers, and S. I. Hay, “Global
kind. transport networks and infectious disease spread,”
Advances in Parasitology, vol. 62, pp. 293–343,
Shape: Morphology extracts picture components. 2006.
Using morphological procedures, diseased fruit and leaf 3. K. A. Garrett, S. P. Dendy, E. E. Frank, M. N.
parts may be removed. The image's borders may be Rouse, and S. E. Travers, “Climate change effects
obtained via erosion. Four major characteristics:
on plant disease: genomes to ecosystems,” Annual
1) Geometric characteristics: Perimeter, area, the Review of Phytopathology, vol. 44, pp. 489– 509,
axis, the orientation angle, and many more terms are 2006.
included. 4. S. M. Coakley, H. Scherm, and S. Chakraborty,
2) Area description features: Region description “Climate change and plant disease management,”
features are provided based on a collection of Annual Review of Phytopathology, vol. 37, no. 1,
characteristics specific to the target area. pp. 399–426, 1999.
3) Moment invariants: Using this feature, geometric 5. S. A. Miller, F. D. Beed, and C. L. Harmon, “Plant
properties like Hu invariant moments, orthogonal disease diagnostic capabilities and networks,”
moments, and so on may be defined. Annual Review of Phytopathology, vol. 47, pp. 15–
38, 2009.
4) Fourier shape descriptor. 6. J. R. Rohr, T. R. Raffel, J. M. Romansic, H.
McCallum,and P. J. Hudson, “Evaluating the links
Step 4 - Classification: Data is properly partitioned using a between climate, disease spread, and amphibian
Support Vector Machine (SVM) to create an N- declines,” Proceedings of the National Academy of
dimensional hyper plane. Neural networks and support Sciences of the United States of America, vol. 105,
vector machines are closely linked. SVM assesses a wider no. 45, pp.17436–17441, 2008.
range of data in an easy-to-understand manner. 7. T. Van der Zwet, “Present worldwide distribution of
fire blight,” in Proceedings of the 9th International
4. CONCLUSIONS Workshop on Fire Blight, vol. 590, Napier, New
Zealand, October 2001.
This research paper describes how to use image processing 8. D. Soni, D. Srivastava "An Empirical Client Cloud
to identify and categorize leaf diseases. To accurately Environment to Secure Data Communication with
diagnose illnesses, writers used numerous techniques. Alert Protocol", Published in Mathematical
 Problems in Engineering, ISSN: 1024-123X. Mining in Healthcare”,2018 International Journal of
9. Dheresh Soni, and M. Kumar. 2019. “An Engineering Technology and Applied Science
Automated Cloud Security Framework Based on (IJETAS) (ISSN: 2395 3853), Vol. 4 Issue 6 June
FCM in User-Cloud Environment.” International 2018.
Journal of Engineering and Advanced Technology 8 14. D. Soni, V. Tiwari, B. Kaur Srao, M. Kumar,
(6). Blue Eyes Intelligence Engineering and “Cloud computing security analysis based on RC6,
Sciences Publication: 3235–40. AES and RSA algorithms”, 2021 1st International
:10.35940/ijeat.F8831.088619. Conference on Advances in computing & future
10. S. Chakraborty, A. V. Tiedemann, and P. S. Teng, communication technologies (ICACFCT - 2021),
“Climate change: potential impact on plant 2021, pp. 1-7.
diseases,” Environmental Pollution, vol. 108, no. 3, 15. I. Steinwart and A. Christmann, Support Vector
pp. 317–326, 2000. Machines, Springer Science &Business Media, New
11. V. Sharma, D. Soni and D. Srivastava, "Filtration York, NY, USA, 2008.
Based Noise Reduction Technique in an Image," 16. H. Cartwright, Ed., Artificial Neural Networks,
2019 4th International Conference on Internet of Humana Press,2015.
Things: Smart Innovation and Usages (IoT-SIU), 17. D. Srivastava, D. Soni, V. Sharma, P. Kumar and A.
2019, pp. 1-6, : 10.1109/IoT-SIU.2019.8777623. K. Singh “An Artificial Intelligence Based
12. M. B. Riley, M. R. Williamson, and O. Maloy, Recommender System to analyse Drug Target
“Plant disease diagnosis. The Plant Health Indication for Drug Repurposing using Linear
Instructor,” 2002. Machine Learning Algorithm” in Journal of
13. Vivek Prajapati and Dheresh Soni “A Review on Algebraic Statistics (ESCI), e-ISSN – 1309-3452, in
Different Application Architecture of Big Data Vol. 13 Issue 3,July 2022.