PRIVACY-PRESERVING IOT-ENABLED PLANT DISEASE CLASSIFICATION WITH AN ENHANCED FEDERATED LEARNING FRAMEWORK AND EXPLAINABLE ARTIFICIAL INTELLIGENCE

Abstract

Plant diseases pose a significant threat to global food security by reducing crop yield and quality in large-scale and geographically distributed agricultural systems, where manual inspection is inefficient and error-prone. Recent advances in Internet of Things (IoT) technologies and deep learning have enabled automated plant disease diagnosis through continuous image-based monitoring. However, centralized learning approaches have issues related to privacy risks, high communication overhead, and limited scalability when handling heterogeneous farm data. To address these challenges, this research introduces a privacy-preserving IoT-driven model for classifying plant diseases using Federated Learning (FL). In the proposed system, leaf images are captured by distributed IoT devices, processed locally, and collaboratively used for FL without sharing raw data, thereby preserving farmer data privacy. Image pre-processing based on resizing and normalization is used to obtain standardized data. Subsequently, a novel FL method is utilized for disease classification with three key contributions. The local model within the FL framework is enhanced by introducing a Hybrid Transformer Network (H-TNet) model. In addition, an adaptive weighted aggregation strategy and loss function optimization using the Chaotic Stellar Oscillation (CStO) algorithm are proposed to improve convergence and enhance the generalization capability of the global model. As a result, the proposed FL model provides a scalable, secure, and accurate solution for real-time plant disease classification.