Spatial-Temporal Graph-Based Distributed Optical Fiber Sensing Method for External Breakage Event Recognition

Improper construction poses a serious threat to the safety of communications cables. Therefore, continuous monitoring along the communication cable is essential to ensure the timely detection of any incidents that may compromise its security. Phase-sensitive optical time-domain reflectometer <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$(\Phi$</tex>-OTDR) has been widely researched in the field of perimeter security due to its high sensitivity, fast response time and multi-point monitoring. However, the environment along communication cables is complex and variable, and long-distance detection often results in time-consuming recognition of external breakage signals and increased susceptibility to misjudgment. To address these issues, we developed a long-distance distributed fiber optic sensing external breakage recognition method based on Spatial-temporal graph. We propose a lightweight RT-DETR model, EFS-DETR, for efficient external breakage detection. The algorithm employs EfficientViT to reconstruct the feature extraction network of RT-DETR, enabling the extraction of deeper semantic features while reducing computational load and model complexity. Secondly, we introduce Fine-Grained Path Augmentation (FGPA) to address the issue of detail feature information loss in multi-level transmission of external breakage signals. Finally, the Shape-IoU loss function is introduced to make the bounding box regression more accurate by focusing on the shape and scale of the bounding box itself to calculate the loss. Additionally, this paper employs a combined threshold iteration and contrast enhancement denoising algorithm for data preprocessing, aiming to reduce various noise interferences while enhancing the external breakage signals of interest. We curated a real-world dataset of external breakage using a 40 km communication fiber optic cable from Guangzhou, China, to validate the efficacy of our proposed method. Multiple sets of experimental results demonstrate that the proposed method surpasses other vibration recognition techniques in fiber optic sensing.