A fiber optic sensing intrusion detection method based on WPD-EMD and improved ResNet

Abstract

Perimeter security systems are essential for safeguarding critical locations from unauthorized intrusions and various security threats. Traditional video surveillance has limitations like coverage gaps, blind spots, and the need for extensive manual analysis. In contrast, fiber optic systems use distributed sensing technology for real-time, precise data acquisition and automatic anomaly detection, reducing the need for manual monitoring. The Phase-Sensitive Optical Time-Domain Reflectometer (Φ-OTDR) is noted for its high sensitivity, large dynamic range, and robust interference resistance, making it ideal for extensive real-time monitoring. To enhance recognition accuracy, especially for high-threat events requiring zero false alarms, this study proposes a fiber optic sensing signal recognition method using Wavelet Packet Decomposition (WPD) combined with Empirical Mode Decomposition (EMD) and an improved ResNet architecture. Encoding one-dimensional signals into images using Recurrence Plots (RP) leverages advanced techniques from image processing and computer vision, enhancing signal recognition accuracy and application scope. Experimental results show that the WPD-EMD denoising method significantly improves the quality of the original signals, achieving an overall recognition rate of 93.83% for eight types of intrusion signals. For the most representative events (background noise, excavator digging, truck passing, stone knocking), the recognition rate reaches 99.69%. This method shows significant potential for advancing perimeter security monitoring.