摘要
为提升相位敏感型光时域反射仪(Φ-OTDR)系统对振动事件径向距离的识别能力,特别是在低信噪比条件下的识别效果。提出一种基于多时频图与注意力机制的识别方法,通过梅尔谱变换(Mel)和离散-连续小波变换(CWT)增强信号时频特征,并构建多通道SE-ResNet50网络,引入通道注意力机制以自适应融合多源特征。实验结果表明,该方法在0~9 m径向距离范围内的相同事件识别准确率达到98.66%,相比单一Mel图、CWT图分别提升5.35%和3.61%,有效增强了Φ-OTDR在复杂环境下的特征表达与识别能力。
To enhance the recognition capability of phase-sensitive optical time-domain reflectometer(Φ-OTDR)systems for identifying the radial distance of vibration events,especially under low signal-to-noise ratio(SNR)conditions,a recognition method based on multi-time-frequency maps and an attention mechanism is proposed.This approach employs Mel-spectrogram transformation and discrete-continuous wavelet transform(CWT)to enhance the time-frequency features of the signal,and constructs a multi-channel SE-ResNet50 network incorporating a channel attention mechanism to adaptively fuse multisource features.Experimental results demonstrate that the proposed method achieves a recognition accuracy of 98.66%within a radial distance range of 0~9 m for identical events,representing improvements of 5.35%and 3.61%compared to using only Mel-spectrograms or CWT maps,respectively.The method effectively enhances the feature representation and recognition performance ofΦ-OTDR in complex environments.
作者
施羿
梁立功
封皓
SHI Yi;LIANG Ligong;FENG Hao(College of Engineering,Shantou University,Shantou Guangdong 515063,China;Xinjiang Haoyuan Water Supply Co.,Ltd.,Hami Xinjiang 839000,China;School of Precision Instruments and Opto-Electronics Engineering,Tianjin University,Tianjin 300072,China)
出处
《光通信技术》
北大核心
2025年第5期78-83,共6页
Optical Communication Technology
基金
国家重点研发计划项目(2022YFC3203802)资助。
关键词
相位敏感型光时域反射仪
时频分析
注意力机制
事件识别
phase-sensitive optical time-domain reflectometer
time-frequency analysis
attention mechanism
event recognition
