Deep Neural Network-Based Approach to Identify the Paddy Leaf Disease using ResNet50-V2

Abstract

South Asian countries especially Bangladesh, India, and Srilanka are very involved in paddy cultivation. In Bangladesh and India, one of the key producing crops is paddy. Worldwide more than 40 percent of the world's population derives most of its calories from rice and traditionally farmers are directly involved production of rice. But every year lot of farmers have to face a tremendous amount of loss due to paddy leaf disease. Major of the farmers do not know how to identify perfectly in the primary stage that's why they are facing loss. Due to the increase in population, it is very important to make paddy cultivation scientific and produce more. There are more than 30 types of paddy leaf disease and from there in Bangladesh, approximately 10 types of disease are quite similar. If it is possible to ensure the leaf disease in the very primary stage within a very short time, then every farmer can reduce their loss. Nowadays many researchers are interested in identifying paddy leaf disease using modern science. In this paper, we worked on paddy leaf disease detection and it has worked on 4 types of disease and 1 healthy and without any disease leaf class of paddy. This paper through an automated detection approach using deep learning techniques achieved the highest accuracy instead of the previous applicable working model and traditional manual disease detection process. In this paper, we have analyzed it using 4 different types of models. ResNet50-V2, Inception-V3, MobileNet-V2, and DensNet-201 algorithm models have been applied here and the finding of the research is, that all the models achieved more than 95% accuracy from these ResNet50-V2 has achieved the best accuracy and it is 99.06%. We hope that these deep learning models and techniques will detect the disease with high accuracy in the very beginning time and it will help reduce damage and loss of paddy cultivation. © 2023 IEEE.