Edge AI-Driven Lightweight Intrusion Detection for Underwater IoT Wireless Sensor Networks: Enhancing Adaptability, Efficiency, and Real-Time Security

Abstract The utilization of Underwater Internet of Things Wireless Sensor Networks (UIoTWSN) is important for the management of resources, monitoring marine environment and conducting environmental evaluations. They dynamic nature of underwater habitats and the limits of Intrusion Detection System are available and provide few obstacles. Because the traditional approach frequently suffers from high computing complexity, raised false positive rates and energy inefficiency, they are not efficiently suited for use in underwater networks that have limited resources. An Edge AI driven Lightweight Intrusion Detection System (Edge-AI IDS) for UIOTWSN is proposed in this study. This method also overcome the issues in the existing methods. The system makes use of method that are based on TinyML such as MobileNetV3 and Gated Recurrent Units for real time detection at edge nodes. Hence, it reduces the amount of processing overhead. Both dynamic transfer learning and meta learning are employed into the system to enhance the adaptability and enables the system to react with evolving threats. To refine decision making, a context aware detection method modifies the sensitivity of the system based in environmental conditions, which reduce the number of false positives. Moreover, techniques that are effective in energy consumption such as quantization and neural network are utilized to conserve power without managing detection accuracy. The decentralized nature of framework uses federated learning and blockchain technology, which ensure the confidentiality of data. It also ensures that network nodes can communicate safely with one another. The result of the experiments shows that the proposed method achieved high accuracy rate and reduced false positive rates in comparison with existing approaches. The proposed method is both scalable and robust for the protection of underwater networks.