Spatiotemporal image-based method for external breakage event recognition in long-distance distributed fiber optic sensing

Abstract

Distributed fiber optic sensing has garnered significant attention in the field of perimeter security. However, existing research seldom addresses the unique challenges posed by long-distance detection, which include two critical issues for data-driven approaches: susceptibility to noise in complex environments and the rapid identification of multiple events within massive datasets. To address these challenges, this study proposes a spatiotemporal image-based method for external breakage recognition using long-distance distributed fiber optic sensing. For data preprocessing, Adaptive Signal Enhancement Denoising (ASED) is proposed to effectively reduce noise in spatiotemporal images converted from raw data. For the recognition model, Lightweight Spatiotemporal Perception Enhanced YOLO (LSPE-YOLO) is proposed, leveraging a one-stage network structure to facilitate fast multi-event recognition in spatiotemporal images. The proposed model integrates a space-to-depth strategy within its backbone network to enhance spatiotemporal awareness with fewer parameters. Additionally, Coordinate Attention (CA) is incorporated into the neck network, which better captures the long-range dependencies of spatiotemporal images. A realistic dataset was constructed by utilizing a 40km communication fiber-optic cable. Experiments show that the method proposed reaches 0.93 on mAP50. The number of parameters and Giga Floating Point of Operations (GFLOPs) are 2.8M and 7.6, respectively. There are obvious advantages in practical engineering applications.