Leaky Cable Perimeter Intrusion Detection Based on Deep Reinforcement Learning

Many existing leaky cable intrusion monitoring methods and systems on the market are based on threshold detection, which can lead to misclassification and omission, limited adaptability to various installation environments, and low localization accuracy. In this article, we present a deep reinforcement learning-based approach for monitoring perimeter intrusion in leaky cables, leveraging the self-adaptation of the interaction between the reinforcement learning agent and its environment. This approach is a typical Internet of Things (IoT) application, integrating advanced machine learning techniques with IoT devices for enhanced security and adaptability. By converting threshold detection into an intelligent classification decision problem and devising a reward function tailored to the perimeter intrusion localization scenario, the learning agent can continually adapt to generate an optimal model suitable for dynamic IoT environments. To assess its feasibility and practicality, experiments in both simulated validation and field deployment environments are conducted. The experimental results indicate that the proposed model achieves 98.73% accuracy for the test set, 1-m accuracy in field localization, a false alarm rate of less than 1%, and robust adaptability to environmental noise, demonstrating its potential as an effective IoT solution for perimeter intrusion monitoring.