Worldwide, rice stands as an essential staple crop with multiple leaf diseases which affect its yield yet remain undetected during initial stages of development. The traditional methods for diagnosing diseases depend on experts who need to conduct manual inspections, which results in slow and inaccurate processes that cannot handle operations of extensive agricultural fields. Deep-learning methods have proven to be highly effective for detecting plant diseases, but these methods need all data to be gathered in one place. The situation creates major problems because it involves protecting data privacy, needing continuous internet access, causing more data transmission since users must provide extra information, and exposing systems to potential cyberattacks. The project develops a Lightweight Federated Learning (LWFL) framework to predict rice leaf disease which allows training on various edge devices without sending actual data to main servers. The proposed system uses a pre-trained InceptionV3 Convolutional Neural Network (CNN) model which undergoes optimization through pruning and data augmentation and image normalization techniques to achieve reduced computational requirements while maintaining precise results. The Flower federated learning framework handles client communication, uses Federated Averaging (FedAvg) to combine model updates, and supports decentralized learning. The dataset is distributed among simulated clients to mimic realworld conditions where each farmer or edge node holds its own local data. The optimized federated model reaches approximately 98.5% accuracy, while showing strong results in both macro and weighted precision and recall and F1-score performance according to the experimental results. The system maintains its capability to perform under conditions of data diversity and resource limitations on client systems. The project demonstration shows that lightweight federated learning models can be used for agricultural disease detection because they provide privacy protection and efficient resource use while offering scalable AI solutions which work in both rural areas and actual farming operations.

The project achieved its goals by developing and implementing a system which used Lightweight Federated Learning (LWFL) to predict rice leaf diseases with accurate results while safeguarding user privacy. The proposed framework achieved its goals by distributing training to multiple clients while maintaining raw image data on individual devices which solved the key drawbacks found in traditional centralized deep-learning models that used deep-learning models which required stable internet access and faced data privacy risks. The Flower FL framework established client-server communication connections which enabled local model updates to be combined and training to continue through non-IID (non-uniform) data conditions. The research team used InceptionV3 CNN model which had been pre-trained to achieve better computing efficiency through pruning techniques and data augmentation and normalization methods which cut trainable parameters without damaging model efficiency. The testing of the centrally reserved test dataset demonstrated outstanding classification accuracy while maintaining high precision and recall and F1-scores across all nine rice leaf disease categories. The confusion matrix analysis confirmed model reliability through its ability to distinguish between visually similar disease classes. The system operational success demonstrates how federated learning can benefit agricultural systems which operate in rural regions that face challenges with data security and resource limitations and restricted internet access. The project establishes a solid base for AIdriven plant disease detection tools to operate in real-world scenarios because it allows on-device training and protects user data. The project shows that Lightweight Federated Learning enables the development of smart farming systems which operate efficiently and securely while maintaining their ability to expand. The system will help farmers achieve better results while decreasing crop losses and adopting environmentally friendly farming methods.

[1]McMahan, H. B., Moore, E., Ramage, D., et al. (2017). Communication-efficient learning of deep networks from decentralized data. *Proceedings of the 20th International Conference on Artificial Intelligence and Statistics (AISTATS)*, pp. 1273–1282. [2]Li, T., Sahu, A. K., Talwalkar, A., Smith, V. (2020). Federated learning: Challenges, methods, and future directions. *IEEE Signal Processing Magazine*, 37(3), 50–60. [3]Kairouz, P., McMahan, H. B., et al. (2021). Advances and open problems in federated learning. *Foundations and Trends in Machine Learning*, 14(1–2), 1–210. [4]Howard, A. G., Zhu, M., Chen, B., et al. (2017). MobileNets: Efficient convolutional neural networks for mobile vision applications. *arXiv preprint arXiv:1704.04861*. [5]Han, S., Mao, H., Dally, W. J. (2015). Deep compression: Compressing deep neural networks with pruning, trained quantization and Huffman coding. *arXiv preprint arXiv:1510.00149*. [6]Reddi, S., Charles, Z., Zaheer, M., et al. (2021). Adaptive federated optimization. *Proceedings of the International Conference on Learning Representations (ICLR)*. [7]Chollet, F. (2017). Xception: Deep learning with depthwise separable convolutions. *Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR)*, pp. 1251–1258. [8]Kingma, D. P., Ba, J. (2015). Adam: A method for stochastic optimization. *International Conference on Learning Representations (ICLR)*.

© 2025 International Journal of Engineering and Techniques (IJET).

Digital Object Identifier (DOI)

IJET assigns a unique DOI to every accepted article via Zenodo for persistent linking and citation. Your article’s DOI: https://doi.org/{{doi}}

Open DOI About Zenodo DOI

Close