An Advanced CNN based System for Rice Plant Diseases Detection and classification in Precision Agriculture

Abstract

For the vast majority of people on the planet, agriculture is one of the most crucial industries. Rice is a vital food source, but it also faces serious risks from illnesses that can affect yield quality and quantity, potentially resulting in 20% to 40% crop losses. Reducing production losses requires early diagnosis of these disorders. For farmers, however, it is not possible to manually monitor such large areas of field. The proposed model based on a Convolutional Neural Network of Resnet50 and transfer learning model, specifically a modified TLResnet152V2. Transfer learning-based weights are used for precise identification and categorization and detection for rice leaf diseases. The proposed system demonstrates high accuracy in identifying four classes of rice diseases viz: Brown spot, Healthy, LeafBlast, NeckBlast. It achieves a remarkable accuracy of Resnet50 & TLResnet152V2 is 93.20% and 95.03 % for normalized augmented dataset respectively. Comparative analysis reveals that modified approach outperforms similar methodologies applied to the Bangladesh rice dataset or datasets of comparable size given in the existing research work. Overall findings underscore the efficacy of proposed approach in advancing the precision of rice plant disease detection, offering a promising solution to enhance agricultural practices and mitigate production losses.