Classification of Tomato Diseases using Hybrid Model (CNN-SVM)
Nishant Garg, Radhika Gupta, Maninder Kaur, Vinay Kukreja, Anuj Kumar Jain, Raj Gaurang Tiwari
Abstract
Tomato cultivation forms the backbone of the vegetable production sector in most countries and is a vital constituent of the staple diet of a variety of populations across the globe. Although multiple factors affect production, plant diseases are the foremost among them and need to be tackled to preserve their worth. This is one of research area of agriculture. The authors seek to pinpoint this fundamental aspect affecting the growth of the plant emphasizing that early detection of plant diseases is necessary to confront their root causes at the initial stages thus, they have deployed a deep learning model constituting CNN and SVM which can recognize 7 prevalent diseases namely Bacterial Spot, Spider Mite, Target Spot, Early Blight, Late Blight, Bacterial Wilt plus Fusarium Wilt and categorize the images into 8 classes comprising of respective diseases as well as the healthy one. The model happened to be trained on a dataset containing 8,000 images of the considered classes and is tested on a test set having 2,000 images. A hybrid approach has been proposed via concatenation of Convolutional Neural Network (CNN) and Support Vector Machine (SVM). CNN is used for the extraction of utilitarian features from the input data and these are then classified using an optimized Support Vector Machine classifier and results are recorded in terms of accuracy. An accuracy of 92.6 % has been demonstrated by the suggested methodology and has thus performed proficiently.