Multi-Band Frequency Interaction Transformer for Perimeter Disturbance Detection Using φ-OTDR

Accurate recognition and localization of disturbance events are essential for reliable perimeter monitoring using phase-sensitive optical time-domain reflectometry (φ-OTDR). An improved YOLOv5s model incorporating a multi-band frequency interaction Transformer (MFIT) is proposed for cross-frequency feature fusion. After components below 5 Hz are suppressed, the disturbance signals are represented by synchronized low-band (5–50 Hz), middle-band (50–100 Hz), and high-band (>100 Hz) spatiotemporal maps. These maps are processed by parallel YOLOv5s branches, with MFIT modules applied at multiple feature scales. Experiments are conducted on seven disturbance classes collected from a cement-fence installation. On the independent test set, the proposed model achieves a precision of 0.984, a recall of 0.993, and an mAP@0.5:0.95 of 0.847. Compared with the Full model, precision, recall, and mAP@0.5:0.95 are improved by 7.1, 6.2, and 9.7 percentage points, respectively. The results demonstrate the effectiveness of multi-band representation and cross-frequency feature fusion for disturbance detection.