A Secure Federated Transfer Learning Approach for Cotton Leaf Disease Classification Using Deep Learning Neural Networks

Cotton, often referred to as ‘white gold,’ plays a vital role in agricultural economies, particularly in Bangladesh, where many livelihoods depend on farming. Despite favourable soil conditions, adequate water resources and a suitable climate, cotton production is frequently reduced by leaf diseases and pest infestations. As most cotton diseases primarily affect plant leaves, early and accurate detection is essential. Conventional visual inspection by farmers or agricultural experts is slow and prone to errors due to the similarity of disease symptoms. To address these challenges, this study proposes a federated transfer learning-based cotton leaf disease detection system that enables collaborative model training across distributed clients without sharing raw image data. Pre-trained deep learning architectures, including VGG-16, VGG-19, Inception-V3 and Xception, are fine-tuned locally on client devices, and their model updates are aggregated using a federated optimization strategy to build a robust global model. This approach preserves data privacy while leveraging the feature learning capability of transfer learning. The experimental results demonstrate that the federated Xception model achieved the highest accuracy of 98.70%, and was deployed in a smart web-based application for privacy-preserving and real-time disease prediction. The proposed system provides reliable cotton disease diagnosis and can be extended to other crops for secure and automated leaf disease detection in distributed agricultural environments.