分布式光纤传感系统的压缩感知和智能传感信息提取技术研究

Brillouin optical time domain sensors were studied intensively and has been improved dramatically in the past few years, how to collect original information efficiently and fast obtain useful information has become the urgent needs of the development of fiber optics distributed sensors. In this study, we propose to utilize compressed sensing, image processing, and signal processing theories and technologies on fiber optics distributed sensors, and we use Brillouin Optical Time Domain Analyzer system as an application example (i.e. CS-BOTDA), the work includes researches on: 1) Sparse sampling of Brillouin spectrum in k-space: appropriate acquisition trajectory and method is employed to acquire only the non-sparse part of the Brillouin spectrum. 2) Reconstruction of the under-sampled Brillouin spectrum: full Brillouin spectrum can be recovered from Orthogonal Matching Pursuit or L1 norm regularization algorithms with the non-sparse spectrum collected, the Brillouin data collection time could be reduced several times with this higher collection efficiency. 3) Intelligent sensing information analyzation and extraction from the spectrum: Image and signal processing methods are put forward to separate the fiber sections carrying sensing information, and then Brillouin shift is instantly obtained with edge detection or other technologies, which cuts the Brillouin data retrieving time to few tenths of times. The successful implementation of this project will not only be a boost to BOTDA systems, the methodology, theories and technologies also can be applied to other fiber optics distributed sensors.

分布式光纤传感系统已研究多年，日趋成熟且结构复杂；如何高效采集原始信息，并从中快速提取有用信息，成为分布式光纤传感系统发展的迫切需求。本项目提出，将压缩感知技术和图像信号处理技术理论用于分布式光纤传感系统，并以布里渊光纤分布传感系统为实际应用场景（即CS-BOTDA），包括：1）布里渊频谱图k空间稀疏采样的研究：选择合适采集轨迹和方式，仅采集布里渊频谱的非稀疏部分；2）布里渊频谱图的欠采样重构研究：利用正交匹配追踪、L1范数约束等算法重构完整布里渊频谱，可数倍降低布里渊数据采集时间、提高采集效率；3）传感信息智能分析提取的研究：利用图像处理手段，分离携带传感信息的光纤区段，并利用边缘提取等算法直接提取布里渊频移，可数十倍降低数据分析时间，快速获取传感信息。本项目的实施不仅有益于BOTDA系统，其方法和技术理论还可以推广应用于其他的分布式光纤传感系统。

在各领域人才的推动下，各种分布式光纤传感系统的系统性能逐年提高，尤其是在空间分辨率、传感距离、测量分辨率和动态范围等参数指标取得了大量突破，且在铁路、电力、石油等得到了广泛的应用，而智能数据处理成为其进一步发展和应用中需要突破的关键。在本项目资助下，负责人团队对Brillouin分布式光纤传感器和Phi-OTDR开展了大量的相关工作，有望在数据处理方面大幅改进分布式光纤传感系统性能。我们的主要研究工作如下：1）完成了基于K-SVD的压缩感知方案，仅需随机采集15%的频率即可恢复整幅增益谱；2）完成了基于各向异性类算法的布里渊增益谱降噪，可高保真地提升SNR约20 dB；3）完成了基于谱减法和基于BP神经网络的传感信息段提取，通过关注频移区段，提高了BFS的准确性和提取速度；4）搭建了一套Phi-OTDR实验系统，针对“各单位的数据集和代码均未公开，难以公平比较算法”的痛点，构建了相关数据集并预计在近期公开；5）利用AdaBoost.M1-KNN-SVM和基于注意力机制的TCN + BiLSTM等网络实现事件分类，在实际数据集上获得93.4%准确率和0%误报率。我们期待本项目相关工作能为从事分布式光纤传感领域研究的研究人员起到抛砖引玉之用。