A secure federated learning framework with blockchain-based authentication for anomaly detection in cooperative smart farming

Abstract

Agriculture is still a crucial global pillar of the world economy, ensuring food security and livelihood. But changing climate conditions and increasing reliance on the Internet of Things (IoT) based precision agriculture have brought new challenges, mainly cybersecurity. Cooperative Smart Farming (CSF) schemes promote resource pooling among small farmers and thus, the adoption of cutting-edge technologies. Although democratizing precision farming also makes it more accessible to cyber threats. To sort out all these problems, the proposed method introduces a secure federated learning framework along with blockchain-based authentication for anomaly detection in CSFs. The proposed method employs a federated transfer learning-based shadow attention mechanism, in conjunction with the Zeiler and Fergus network, to amplify the performance of the anomaly detection technique. This allows farms to detect threats in premises while protecting data privacy, as only model updates that are encrypted are transmitted. A blockchain-backed decentralized authentication scheme based on the proof of authentication consensus protocol is built to provide trust and resistance to tampering. Moreover, the random k-sparsification technique with changing rank adjustment has a positive impact on communication efficiency, which results in the same level of accuracy but with less communication overhead. The proposed method is executed on four datasets, CSE-CIS-IDS2018, MQTTset, ToN-IoT, and Edge IoT, achieving classification accuracies of 99.99%,99.93%,99.94%, and 99.91%, respectively, and it performs better than the current techniques. This comprehensive framework provides a scalable and trusted method for anomaly detection in CSFs, enabling resilient, privacy-preserving, and intelligent smart farming ecosystems.