Amandeep Kaur
Chitkara University Institute of Engineering and Technology, Chitkara University, Punjab, India.

Kalpna Guleria
Chitkara University Institute of Engineering and Technology, Chitkara University, Punjab, India.

DOI https://doi.org/10.33889/IJMEMS.2026.11.4.074

  ;

Abstract

The rice crop is one of the major food crops in India and around the world, which globally necessitates food security. However, the productivity of rice is usually decreased due to various rice leaf diseases. Nowadays, the identification of such diseases has become increasingly feasible through an automated process, which has been made possible due to various improvements and advancements in deep learning technologies. Traditional deep learning methods are heavily reliant on centralized data collection, thus causing significant issues regarding data privacy and ownership for the agricultural stakeholders. Centralized deep learning models require a large memory space and high computation cost for datasets to be stored at a centralized location. Federated learning solves these problems by allowing collaborative model training in a decentralized way, thus no data needs to be shared with a centralized location. This research proposes FedLeaf, a collaborative federated deep learning model for the accurate identification of rice leaf diseases. The FedLeaf design adopts a distributed learning method in which the global model is forwarded to client nodes, and then clients apply local training using their respective datasets. In the proposed research, FedLeaf utilises two deep learning architectures, MobileNetV2 and EfficientNet, in a federated environment, which are named as federated learning based MobileNetV2 and federated learning based EfficientNet. The hyperparameter tuning has been performed to get the best results, at a batch size of 32, learning rate of 0.001, Adam optimizer, seven communication rounds and five epochs per communication round. The performance evaluation of the proposed FedLeaf, has been performed based on its round-wise global and local performance parameters, including accuracy, precision, recall, F1-Score and AUC. In the proposed FedLeaf model, federated learning based EfficientNet has emerged as the optimal model for rice leaf disease detection, exhibiting 96.19% highest accuracy for IID distribution and 95.87% for non-IID data distribution. The results exhibit that the proposed model shows better performance and offers a collaborative and efficient model for automated identification of rice leaf disease.

Keywords- Federated learning, Rice leaf disease, EfficientNet, MobileNetV2, Precision agriculture, FedLeaf.

Citation

Kaur, A., & Guleria, K. (2026). FedLeaf: A Collaborative Federated Learning based Model for Rice Leaf Disease Detection. International Journal of Mathematical, Engineering and Management Sciences, 11(4), 1830-1862. https://doi.org/10.33889/IJMEMS.2026.11.4.074.