随机数编码融合图像特征提取提升BOTDR应变检测性能研究

基于布里渊光时域反射仪(BOTDR)的应变检测是长距离结构健康监测的重要手段，然而其传感距离、空间分辨率、测量精度和响应速度之间彼此制约，综合性能提升受限。研究在不降低空间分辨率的前提下同时提升长距离BOTDR信噪比和测量速度的新理论，发展长距离、高精度、快速检测BOTDR新方法，突破BOTDR综合性能提升瓶颈是重要的科学问题。.本项目提出用随机数序列调制探测脉冲的方法，研究随机数编码的布里渊后向散射时域特征和频域分布特性，建立随机脉冲编码探测信噪比提升理论，确定编码增益与码长关系，并通过实验验证确定最佳码长，实现高编码增益下的信噪比提升和非线性效应抑制，在不降低空间分辨率的前提下提升传感距离；融合深度卷积神经网络实现高精度快速布里渊增益谱三维图像特征分析，并提取应变、温度同时改变下的布里渊增益谱强度、频率与谱宽特征差异，消除温度对应变测量的串扰，最终提升BOTDR应变检测性能。

The optical fiber strain detection based on Brillouin Optical Time Domain Reflectometer (BOTDR) is an important mean of health monitoring and risk early warning for long-distance infrastructure structures. However, its sensing distance, spatial resolution, measurement accuracy and response speed are constrained each other, and its comprehensive performance is limited, affecting fine detection. On the premise of not reducing spatial resolution, it is an important scientific problem to improve the signal-to-noise ratio and measurement speed of long-distance BOTDR, develop new long-distance, high-precision and fast detection method, and enhance the comprehensive performance of BOTDR..This project demonstrated a new scheme of BOTDR incorporating random number coding and image feature extraction. The Brillouin time domain signal and frequency domain distribution of random-number-coding BOTDR are analyzed, based on which the theory of SNR enhancement by random number coding is built. Through exploring the optimal bit length by numerical simulation and experimental verification, the enough coding gain can be achieved and the nonlinear effect can be restrained. Then, the sensing distance is extended without deterioration of spatial resolution. Furthermore, the three-dimensional image features are analyzed utilizing deep convolutional neural network (DCNN). And the temperature and strain can be distinguished by extracting characteristic differences of Brillouin gain spectrums which are obtained when temperature and strain are changed simultaneously. Eventually, the overall performance for strain detection based on BOTDR is improved.