Detection and Categorization of Rice Leaf Diseases through Federated Learning and Improved Vision Transformer Models

Rice (Oryza sativa L.) is an important food source for people worldwide. In Asia, where it is mostly grown and eaten, it provides 14% of protein and 22% of calories per person. Microbial diseases like viral, bacterial, fungal, and other illnesses are bad for health and food production, which is a big problem for rice farmers. It is hard to tell if these diseases are present physically, especially in places that do not have crop safety specialists. Managing disease detection and making decision-making tools easy is important for making Rice Leaf (RL) protection techniques work and lowering damage to rice crops. Unfortunately, there is not yet a reliable and safe way to diagnose RL disease, even though many options exist. Federated Learning (FL) is an appealing and effective way to deal with these issues. The study suggested using FL and improved Vision Transformers (VT) models to find and classify RL diseases. Highly specified transfer learning (TL) models and the suggested design that uses the Self-Attention Mechanism (SAM) have been tested. These models are then combined into a decentralized learning method based on FL. The suggested architecture utilizes the advantageous interactions of VT models, CoAtNets, and the improved Swin Transformer (ST) V2, leading to a superior representation of features. The suggested model in the FL system markedly surpasses all previously evaluated TL models, attaining accuracies of 99% for RL disease categorization.