Optimization Based Rice Leaf Disease Classification in Federated Learning

Abstract

Numerous farmers worldwide are impacted by diseases connected to rice leaves that frequently endanger the sustainability of the rice industry. Diseases that affect the leaves of rice plants severely limit their ability to produce rice, and they are typically brought on by bacteria, viruses, or fungi. This paper proposes an innovative classification scheme for rice leaf diseases based on Federated Learning (FL). Here, FL framework comprises two entities, namely nodes and servers. Every node does initial local training using local data. Moreover, produced local model is then updated on server. Model aggregation is done at the server since several nodes update their local models and send them to it. The nodes download the global model that server has generated as a result. The nodes update their training using transferred global model and local model. The following series of actions are taken in the training model. The input image is mainly obtained from a database and pre-processed with a Kalman filter to eliminate noise. Then, numerous operations for data augmentation are applied. In addition, feature extraction is done and generated features are used in LeNet for rice leaf diseases classification. LeNet is trained using the Spotted Hyena Archimedes Optimizer (SHAO). The developed method shows better precision of 91.3%, recall of 92.2%, f-measure of 91.7%, loss function of 3.3%, Mean Square Error (MSE) of 7.3%, and Root MSE of 27.1%.