Image Processing Techniques for Plant Disease
Detection
Article ID: 40830
Revati Ramesh Nalawade1
1Ph. D. Scholar, Department of Plant Pathology, College of Agriculture, Dr. Balasaheb Sawant Konkan
Krishi Vidyapeeth, Dapoli, Ratnagiri, Maharashtra – 415712.
Introduction
During these past few years, Agriculture has become an important source of economic development in
India. To meet the increasing demands of Indian population, agricultural industries are continuously
searching for new advanced techniques to increase the farm production with reduced input cost to aid the
economic development of the country. This encourages researchers to develop new improved and efficient
technologies to increase productivity. For successful farming system timely disease management is very
important to minimize yield losses. If a farmer wants to check for any plant disease, he goes for naked eye
observation of the plant or expert advice after surveying the field, which is very time consuming and
expensive. This may lead to delay in control measure application which will ultimately result in increased
yield losses. This problem can be solved using image processing techniques for plant disease detection.
Digital image processing detects plant disease at an earlier stage than the human eye could recognize them.
This technique involves image acquisition, image pre-processing, image segmentation, feature extraction
and classification of disease. It enables the farmers to take timely disease control measures to minimize
the losses and improve the produce quality. In this paper, the steps of image processing techniques have
been discussed along with different classifiers used for plant disease classification.
Steps Involved in Development of General Plant Disease Detection System Using Image
Processing Techniques
Plant diseases are identified by observing different plant parts for disease symptoms. Image processing
techniques can be used for detection of leaf flower, fruit, stem and root diseases also. Image processing
techniques involve following steps (Fig.1):
Image acquisition: First step for developing any image processing system is image acquisition. The image
acquisition is mostly carried out in real time in controlled condition or under field conditions. High quality
images can be collected using drones, digital camera, smartphones or scanners. These images are used as
input data for the image processing model, the input image data should be in .bmp., .jpg., .png., or .gif
format. Minimum 1000 images in any mentioned format are required for analysis. To ensure accuracy of
data image collection in controlled environment under even light conditions is mostly preferred.
Dataset Annotation: The collected image dataset is annotated for name, date, time, plant type, plant part
type or disease type for knowledge-based dataset creation.
Image pre-processing: To improve the image data features different image pre-processing techniques
are applied, also called as image restoration. This step involves image cropping, image resize, shape
adjustment, image smoothing, contrast and brightness adjustment, colour conversion and noise removal to
highlight the diseased area from an image. Before actual image processing unrequired data and objects are
removed form the image.
Image segmentation: partitioning an image into different parts of same features is called as image
segmentation. Image segmentation is done using different techniques like Otsu’s, k-means clustering,
thresholding, region and edge-based methods etc. it is one of the difficult tasks in image processing and
first step in image analysis and pattern recognition. This step simplifies the image into more meaningful
form for easier analysis. Specific segments of specific characters in an image which represents the
particular symptom area can be segmented in more advanced steps.
Volume 05 - Issue 05 - May 2023 90 | P a g e
Feature extraction: It is the most important part of disease classification system. In this step image
features like colour, shape, texture and morphology are extracted using different feature extraction
techniques. How the colour is distributed in the image the roughness, the hardness of the image is called
as texture of the image. For detection of plant leaf diseases, morphology feature extraction is better than
colour and texture feature extraction. Most commonly used feature extraction techniques involve Grey level
Co-Occurrence Matrix (GLCM), Global Color Histogram (GCH), Blend vision and Machine intelligence etc.
Disease Classification: the extracted features are given different classes like diseases and healthy which
are used to train the Machine learning or Artificial intelligence model to classify the remaining images into
given classes using the features. It is the most challenging task in image processing.
Diagnosis: after all these steps are completed and the exact differentiating features set is created the
model is trained using 1000-2000 images. Then is it field validated and upgraded with each field validation
until it achieves 90-100% accuracy. Most of the models show 60-70% accuracy in first field validation.
Fig. 1 Steps in Image Processing
Some Commonly Used Classifiers for Plant Disease Detection
1. Support Vector Machine: SVM is a supervised types of learning algorithm based on structural risk
minimization. It is mostly used for classification and regression problems. This classifier is designed to
maximize the classification boundaries between two classes as widely as possible.
2. Convolutional Neural Network (CNN): CNN is a class of deep forward neural network that processes
multidimensional data. CNN reduces the image into an easier-to-process form essential for good prediction
without reducing image features. CNN is available in different architectures such as ResNet, VGGNet,
AlexNet, GoogLeNet etc. CNN model consists of an input layer, convolutional layer, max pooling layer, a
fully connected layer and an output layer. The diseased plant images are provided as input then CNN
extracts required features from the images with the help of convolutional and pooling layer to obtain more
details. The output from these layers is transformed into a single vector by fully connected layer which is
used as an input for next layer. Finally, the output layer classifies the plant disease.
3. K- Nearest Neighbor (K-NN): This is a statistical and non-parametric classification system where the
weight is given corresponding to neighbors. The classification is done based on the computed Euclidean
distance metric. It stores all the training tuples given to it as inputs in its learning phase without doing
any calculations hence, called as lazy learner. This prevents its uses in areas where dynamic classification
for large databases is needed. This technique is widely used for text mining, pattern recognition, forecasting
the trends in stock market and plant disease classification in agriculture.
Volume 05 - Issue 05 - May 2023 91 | P a g e
4. Artificial Neural Network (ANN): ANN is an information processing system which works like the
biological system i.e., brain. It consists of interconnected artificial neurons of processing elements which
form neural structure. They gather information by recognising data patterns and relationships and learn
through experience and not by programming. Artificial neurons consist of several inputs that take any
value between 0 and 1, but only a single output. Due to their capability of deriving meaning from complex
database they are mostly used for pattern recognition. Feed forward ANNs and Feedback ANNs are the
two types of ANNs. In Feed forward ANN, the behavior of any layer will not affect that same layer. In
Feedback ANN, signals communicated in both directions by network loops.
Conclusion
Several plant diseases lead to annual yield losses in agriculture. Therefore, plant disease detection at an
early stage is very crucial to minimize yield losses as well as reduce the input and control measures cost.
In this paper, advanced image processing techniques have been discussed which are very efficient as
compared to tradition plant disease control methods. It can be concluded that with further modifications
and increased on field real time validation these image processing techniques can achieve accuracy up to
90-95% which will help for early detection and management of plant diseases ultimately reducing yield
losses in agriculture.
References
1. Kartikeyan P. and Shrivastava Gyanesh, (2021). Review on Emerging Trends in Detection of Plant Diseases using Image
processing with Machine Learning. Int. Jr. Com. App. 174(11):39-48.
2. Sharma V., Verma S. and Goel N., (2020). Classification Techniques for Plant Disease Detection. IJRTE. 8(6):5423-5430.
3. Shruthi U., Nagaveni V. and Raghavendra B. K., (2019). A Review on Machine Learning Classification techniques for Plant
Disease Detection. Proceedings of 5th International Conference on Advanced Computing and Communication systems. Pg no.
281-284.
Volume 05 - Issue 05 - May 2023 92 | P a g e
