Strengthening Smart Irrigation System Against Network Attacks

Abstract

The integration of IoT technologies in smart irrigation systems has improved agri-cultural efficiency through real-time monitoring and automated water management, but exposes these systems to cyber threats such as DoS, DDoS, spoofing, and flood-ing attacks, which can disrupt irrigation operations and cause severe crop damage. To address this, we propose DeepIrri-IDS, a deep learning-based intrusion detection system deployed at the fog layer, that integrates SimPy- based irrigation simulation, cyber-attack injection using the Farm-Flows dataset, and SMOTE oversampling to handle class imbalance, evaluating four deep learning architectures: CNN, LSTM, FT-Transformer, and Mamba. Experimental results demonstrate that all models ex-ceed 99% accuracy on the Farm-Flows dataset, with the LSTM model selected as the most suitable architecture for real-time deployment, achieving 99.39% accuracy and the fastest inference time of 27.97 µs per packet, offering the best balance between detection performance and computational efficiency in resource-constrained IoT environments, successfully blocking malicious traffic before it reaches the irrigation controller and maintaining stable and continuous system operation despite the presence of diverse cyberattacks.