GuavaCareNet: Empowering Guava Farms with Federated Learning for Real-Time Disease Diagnosis

Abstract

This research presents a Federated Learning-based approach for the detection of guava leaf diseases, aimed at enhancing agricultural practices while ensuring data privacy. Guava farms across various regions provide images of guava leaves, both healthy and diseased, to train a deep learning model for disease classification. The EfficientNet-B3 model, optimized with Neural Architecture Search (NAS) and Transfer Learning, is used for accurate disease detection, including conditions like Anthracnose and Powdery Mildew. To ensure privacy, Federated Learning (FL) is employed, where the model is trained locally on edge devices such as smartphones and Jetson Nano devices situated at guava farms. These devices process the data locally and send model updates, rather than raw data, to a central server. This method enables the aggregation of knowledge from multiple farms without sharing sensitive information, thereby protecting farmers' data privacy. The use of Federated Learning allows the model to adapt to specific conditions at each farm, such as different diseases, leaf appearances, and lighting variations, improving its robustness and accuracy. Additionally, AWS IoT integration allows real-time monitoring and alerts, further empowering farmers with timely disease detection. The decentralized nature of this approach not only enhances disease detection in guava crops but also fosters data privacy, ensuring farmers' information remains secure.