Scholarlink

Phase-sensitive Optical Time Domain Reflectometry( \(\varphi \) -OTDR), as an advanced distributed fiber optic sensing technology, has opened new vistas in home security monitoring. However, the limited event recognition capability remains a key constraint in its practical applications.This paper innovatively proposes an instance segmentation method that combines \(\varphi \) -OTDR technology with the improved YOLOv8 algorithm, named FAC-YOLO (FasterNet-GCA-CARAFE You Only Look Once). The model introduces the Global Coordinate Attention (GCA) module to deeply analyze spatial relationships for precise dynamic feature weighting and employs the Content-Aware Reassembly of Features (CARAFE) strategy to replace traditional up-sampling techniques, significantly enhancing the accuracy of feature reconstruction and detail preservation. Furthermore, substituting the original backbone network by the FasterNet framework resulted in a substantial reduction in model parameters and optimization of computational efficiency, adapting to the computational and storage limitations of miniature hardware. Experimental results demonstrate that the innovatively proposed FAC-YOLO model achieves a precision of 97.4 \(\%\) in detecting home security events, reducing the parameter amount by 13.65 \(\%\) and increasing mAP@0.5 by 1.8 \(\%\) compared to the baseline YOLOv8 model.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Distributed Fiber Acoustic Sensing Home Anomaly Detection Technology Based on Lightweight YOLO

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Shan Chun,
  • Yu Yu,
  • Nan Zhao,
  • Hongming Liu,
  • Qinnan Zhang,
  • Yewen Huang

摘要

Phase-sensitive Optical Time Domain Reflectometry( \(\varphi \) -OTDR), as an advanced distributed fiber optic sensing technology, has opened new vistas in home security monitoring. However, the limited event recognition capability remains a key constraint in its practical applications.This paper innovatively proposes an instance segmentation method that combines \(\varphi \) -OTDR technology with the improved YOLOv8 algorithm, named FAC-YOLO (FasterNet-GCA-CARAFE You Only Look Once). The model introduces the Global Coordinate Attention (GCA) module to deeply analyze spatial relationships for precise dynamic feature weighting and employs the Content-Aware Reassembly of Features (CARAFE) strategy to replace traditional up-sampling techniques, significantly enhancing the accuracy of feature reconstruction and detail preservation. Furthermore, substituting the original backbone network by the FasterNet framework resulted in a substantial reduction in model parameters and optimization of computational efficiency, adapting to the computational and storage limitations of miniature hardware. Experimental results demonstrate that the innovatively proposed FAC-YOLO model achieves a precision of 97.4 \(\%\) in detecting home security events, reducing the parameter amount by 13.65 \(\%\) and increasing mAP@0.5 by 1.8 \(\%\) compared to the baseline YOLOv8 model.