Decentralized Deep Learning for Rose Leaf Disease Detection: A Federated CNN Approach

Abstract

Rose is considered useful in agriculture and horticulture. Their leaves are however highly prone to illness such like black spot, mildew and mosaic virus. Such diseases may reduce the yield as well as quality. The conventional methods of detection are slow and they might require specialist expertise. This renders them not practical to farmers particularly in the countryside. Yesterday our project is aimed to create a smart solution that would keep all data confidential and apply deep learning and Federated Learning (FL) to effectively and safely identify rose leaf diseases. The models employed by the system to make the system fast and accurate include VGG19, Effective Net-v2, and MobileNet V2. The farmers are able to take leaf images or upload them using their mobile phone. Training is actually done local to the device the raw data is never transferred outside of the device. The learning results are collected by a central server and this makes the model scalable, reliable and able to be applicable in other farms. It is a system that can be used on mobile devices and low-power edge devices. It is convenient, simple to use and useful in real world farming. The collaboration between AI, privacy, and usability aims to assist farmers to discover the diseases at their early stages, secure the quality of crops, and create more sustainable agriculture.