Enhancing Tomato Leaf Disease Detection Using Federated Learning for Efficiency and Privacy Preservation

Tomato cultivation is vital for global food security, yet it faces considerable challenges from foliar diseases that can profoundly reduce crop yields. Early and real-time detection of these diseases is essential to maintain agricultural productivity and ensure food security. The proposed method leverages federated learning to enable decentralized training across multiple devices, enhancing data privacy while minimizing communication overhead. The effectiveness of the designed SimpleNetAugDR- 3 model is demonstrated through extensive experimentation. The model achieves a favorable trade-off between classification accuracy, storage requirements, and inference speed. The fine-tuned SimpleNetAugDR-3 model achieved a classification accuracy of 92.28%, with an inference time of 0.672 ms and a compact size of 2.61 MB. These characteristics make it suitable for resource-constrained environments. The optimized model was then evaluated within a federated learning setting using four client subsets, achieving an overall accuracy above 92%. The results highlight the potential of federated learning in providing an efficient and privacy-preserving tomato leaf disease detection model. The source code and data used in this study are publicly available at https://github.com/ons-loukil/tomato-fldisease-detection .