Efficient federated transfer learning-based network anomaly detection for cooperative smart farming infrastructure

Precision agriculture has emerged as a vital solution to meet the food demands of the growing global population. However, the high upfront costs of sensors, data analytics tools, and automation often pose challenges for small-scale farms, limiting their ability to adopt these advanced practices. Cooperative Smart Farming (CSF) provides a practical solution to address the evolving needs of modern farming, making precision agriculture more accessible and affordable for small-scale farms. These cooperatives are formal enterprises collectively financed, managed, and operated by member farms, working together for shared benefits. Though it benefits small-scale farmers, all member farms can embrace advanced technologies through collective investment and data sharing by joining cooperatives. As Smart Agriculture grows, CSFs are poised to be essential in building a more sustainable, resilient, and profitable agriculture for all member farms. However, CSFs face increased cybersecurity risks as technology reliance grows. Cyberattacks on one farm can disrupt the entire network, threatening data integrity and decision-making. Federated Learning (FL)-based anomaly detection has been proposed to address this, allowing farms to detect threats locally and share only model updates. However, cooperatives' data sharing and interconnected nature introduce challenges in developing the anomaly detection model. This model must detect threats early and take preventive actions, as delays could result in successful attacks on other smart farms in the network. Additionally, if more smart farms join the cooperative, the model gradient updates can still be transmitted to the server quickly without overwhelming communication channels and causing delays. To address these challenges, in this research, we develop an efficient Federated Transfer Learning FTL based network anomaly detection model for the CSF environment. We also use a dynamic low-rank compression algorithm to reduce the communication latency. To evaluate this proposed approach, we first set up two independent smart farming testbeds incorporating various sensors commonly used in smart farming. We then launch different cyberattacks in each smart farm and collected two network datasets. For proof of concept, we implement and assess the robustness of our proposed model based on metrics such as identifying anomalies, memory consumption, training time, and accuracy using two network datasets. The experiments demonstrated that our proposed model achieves higher accuracy and requires less training time than traditional FL algorithms, enabling early and efficient attack detection in CSF and minimizing the impact of cyberattacks on member farms.