SECURING LOW-POWER EDGE AI: A VULNERABILITY ANALYSIS AND CYBERSECURITY FRAMEWORK FOR RESOURCE-CONSTRAINED DEVICES

Abstract

The rapid deployment of Edge AI systems powered by low-power technology has created new operational challenges in healthcare facilities, autonomous vehicles, and industrial IoT devices, leading to increased security vulnerabilities. Security mechanisms require strong computational power, which these efficiency and real-time-oriented systems do not possess by design. Predictable threats against Edge AI systems are now more prevalent due to their limited computing power. Security frameworks prove inefficient when used in Edge AI environments, creating a vital protection weakness. This research paper focuses on a new lightweight cybersecurity system designed for Edge AI systems that lack sufficient resources. The proposed security construct enables the deployment of effective cryptographic systems and AI-based anomaly identification along with defence mechanisms suitable for real-time edge devices operating with restricted power consumption.  A realistic dataset, known as the Cyber Threat Detection Dataset, was utilised to test the framework, which incorporated multiple normal and attack behaviours, as well as several different attack varieties. The high-speed intrusion detection system relies on Random Forest (RF), while a Lightweight Convolutional Neural Network (CNN) handles anomaly detection activities within the framework, and Federated Learning (FL) decentralises learning across different edge nodes with privacy protection. According to test results, the developed framework delivers advanced threat detection accuracy, better energy efficiency, and faster inference speeds. Each security model partakes specific capabilities that enhance a layered defence, resulting in a protected, adaptable and scalable protection for Edge AI systems.