Deep Learning Approach Towards Plant Disease Detection

Abstract

Disease detection in plants is where the domains of technology and food safety converge as it impacts both India’s agriculture-based economy and the daily livelihood of millions of Indian farmers. Plant disease can cause annual losses of up to 30%. Therefore, disease detection before crop loss is no longer a choice, it has become a necessity. Traditional Machine Learning approaches have improved significantly in classifying disease; however, Deep Learning (DL) has provided an innovative approach in providing more precise predictions of plant diseases. This research aims to provide an intelligent, and an optimized Convolutional Neural Network (CNN) multi-class plant disease detection framework that works under realistic farming conditions. In the present study, the system uses high-resolution (40 MP) leaf images and applies feature engineering to achieve the best results possible for three economically valuable crops in the Delhi-NCR area: Tomato, Wheat and Mustard. The findings revealed that multi-class disease detection in Wheat achieved 99.40% accuracy while multi-class disease detection in Tomatoes achieved 95.90%. The results far exceeded all benchmarked results for the mentioned crops in practical application scenarios. This framework would allow farmers to access affordable technology which would enable them to detect diseases in their crops earlier than ever before which will help reduce the need for chemical intervention and will allow farmers to act quicker on the crops, they produce thereby increasing the overall resiliency of the food supply chain.