LIGHTWEIGHT DEEP LEARNING IDS FOR ENERGY-CONSTRAINED WIRELESS SENSOR NETWORKS

Abstract

Wireless Sensor Networks (WSNs) are widely used in applications such as environmental monitoring, industrial automation, healthcare, smart cities, and agriculture. These networks consist of small sensor nodes with limited energy, memory, and computational capacity. Due to these limitations and the use of open wireless communication channels, WSNs are vulnerable to security attacks such as Denial of Service (DoS), sinkhole, blackhole, spoofing, and selective forwarding attacks [1], [15]. Traditional Intrusion Detection Systems (IDS) often require high computational power and communication resources, making them unsuitable for energy-constrained sensor networks [2]. Lightweight Deep Learning-based IDS offers a practical solution by using compact neural network models and energy-efficient detection techniques. The proposed approach uses a hierarchical detection strategy in which sensor nodes perform initial anomaly detection using ultra-lightweight models, while cluster heads perform more detailed classification. Techniques such as model pruning, quantization, feature selection, and knowledge distillation help reduce memory usage and energy consumption [4], [6], [16]. The system aims to achieve high intrusion detection accuracy while minimizing energy consumption and communication overhead, making intelligent security more suitable for resource-constrained WSN environments [8], [18].