Comparative Analytics on Chilli Plant Disease using Machine Learning Techniques

This thesis concerns the detection of diseases in chilli plants using machine learning techniques. Three algorithms, viz., Convolutional Neural Network (CNN), Recurrent Neural Network (RNN), and Multi-Layer Perceptron (MLP), and their variants have been employed. Chilli-producing countries, India, Mexico, China, Indonesia, Spain, the United States, and Turkey. India has the world’s largest chilli production of about 49% (according to 2020). Andhra Pradesh (Guntur) is the largest market in India, where their varieties are more popular for pungency and color. This study classifies five kinds of diseases that affect the chilli, namely, leaf spot, whitefly, yellowish, healthy, and leaf curl. A comparison among deep learning techniques CNN, RNN, MLP, and their variants to detect the chilli plant disease. 400 images are taken from the Kaggle dataset, classified into five classes, and used for further analytics. Each image is analyzed with CNN (with three variants), RNN (with three variants), and MLP (with two variants). Comparative analytics shows that the higher number of epochs implies a higher execution time and vice versa for lower values. The research implies that MLP-1 (36.08 in seconds) technique is the fastest, requiring 15 epochs. More hidden layers imply higher execution time. This research implies that the MLP-1 technique yields the lowest number of hidden layers. Thereby giving the highest execution time (349.1 in seconds) for RNN-3. Lastly, RNN and MLP have the highest accuracy of 80% (for all variants). The inferences are that these approaches could be used for disease management in terms of the use of proper pesticides in the right quantity using proper spraying techniques. Based on these conclusions, an agricultural scientist can propose a set of right regulations and guidelines