An Optimized Rice Leaf Disease Classification using Transfer Learning and Balanced Class Weight Distribution based on Bandit Approach

Abstract

Rice crops have been the leading staple food, particularly in South-East Asia. During planting, rice leaves suffered from diseases such as Rice Blast, Hispa, and Brown spot. These diseases may lead to failure in harvesting. In this research, we proposed a method based on transfer learning to tackle such issues. Our method contains several steps. In the first step, the rice leaf is preprocessed. Second, due to data imbalance, balanced class weighting was employed. Third, to improve the network performance, three layers of convolution were added to the transfer learning model. The parameters in fully connected layers were optimized using bandit based approach. In the last step, the leaf was classified into nine categories. We compare our method with the state-of-the-art (SOTA) works. Our model reaches the top in terms of accuracy with 98 % compared to the other SOTA.