Real-Time Φ-OTDR Vibration Event Recognition Based on Image Target Detection

Area of Science:

• Optoelectronics
• Computer Vision
• Signal Processing

Background:

• Accurate identification of vibration signals is critical for distributed vibration warning systems.
• Reducing false alarms in long-distance monitoring is a key challenge.
• Phase-sensitive optical time-domain reflectometry (Φ-OTDR) is a core technology for such systems.

Purpose of the Study:

• To develop a real-time, computer vision-based method for detecting multiple vibration events using Φ-OTDR.
• To enhance the accuracy and efficiency of perimeter intrusion detection systems.
• To reduce personnel patrol costs through automated monitoring.

Main Methods:

• Vibration signal feature enhancement using pulse accumulation, pulse cancellers, median filter, and pseudo-color processing.
• Generation of vibration spatio-temporal images to create a customized dataset.
• Training and evaluation of an improved YOLO-A30 model for target detection.

Main Results:

• Achieved a mean Average Precision (mAP@.5) of 99.5% on a dataset of 8069 vibration images.
• Demonstrated a processing speed of 555 frames per second (FPS).
• Capable of detecting events over a theoretical maximum distance of 135.1 km per second.

Conclusions:

• The proposed computer vision method effectively identifies abnormal vibration activities in real-time.
• The system significantly reduces the false-alarm rate for long-distance, multi-vibration events along high-speed rail lines.
• The method maintains high accuracy while substantially reducing computational cost.