Chapter 11 - Groundnut leaves and their disease, deficiency, and toxicity classification using a machine learning approach

Abstract

In the agricultural field, automatic identification of plant leaf diseases and deficiencies is highly desirable. The proposed work is to automate identification of crop leaves initially and then identification of multiple diseases, deficiency, and toxicity in the plant leaves. The work is validated on a real-time dataset and for crop leaf images available in PlantVillage datasets. The real-time dataset of groundnut leaves are collected from the Maruthakulam area, Tirunelveli, Tamil Nadu. The dataset contains both healthy and infected leaf images. The classification of images is carried out using machine learning approaches. Preprocessing techniques, such as histogram equalization and median filtering, are performed. Features such as color, shape, size, and texture are used for classification. The color-based segmentation is carried out to segment the diseased/affected part from the leaf. The segmented image is used for feature extraction. The features extracted are color, shape, size, and texture, and those features are used for classification using various classifiers. Initially, the type of crop, such as corn, bell pepper, paddy, pomegranate, potato, tomato, and groundnut, are identified using leaves to check the robustness of the proposed method. Crop type is identified by using classifiers, and classification accuracy of about 100% is obtained by the classifiers, such as support vector machines (SVMs), decision trees (DTs), and k-nearest neighbor (KNN), and 99.8% by random forest (RF) and neural network (NN) classifiers. Subsequently classification was carried out in identification of diseases, deficiencies, and toxicity for different crops. For different sets of training and testing samples of datasets containing 20 samples in each category, the results reveals that the classification accuracy yielded by the classifiers in classifying the diseases, deficiencies, and toxicities are 99.84% by SVM, 99.44% by NN, 99.88%, by RF, 99.9% by DT and KNN.