基于深度学习的铁路路基雷达检测信号中强干扰压制方法研究  

Clutters suppression in GPR signal for railway subgrade detection based on deep learning

作　　者：林皓 肖建平[1,2] 刘志航 刘郅静 邓宇[1,2] LIN Hao;XIAO JianPing;LIU ZhiHang;LIU ZhiJing;DENG Yu(Key Laboratory of Metallogenic Prediction of NonFerrous Metals and Geological Environment Monitor,Ministry of Education,Central South University,Changsha 410083,China;Institute of Applied Geophysics,School of Geosciences and Infophysics,Central South University,Changsha 410083,China)

机构地区：[1]中南大学有色金属成矿预测与地质环境监测教育部重点实验室,长沙410083 [2]中南大学地球科学与信息物理学院,长沙410083

出　　处：《地球物理学进展》2023年第6期2714-2723,共10页Progress in Geophysics

基　　金：国家自然科学基金项目(42074171)资助。

摘　　要：探地雷达具有快速、非接触式、无损的特点,已成为铁路路基病害检测的重要手段.但由于铁路路基检测对于探地雷达而言具有复杂的电磁干扰环境,尤其是来自轨枕强反射干扰湮没了路基介质层病害的弱反射雷达信号.针对上述问题,本文开展了铁路路基雷达检测信号中轨枕强干扰压制方法研究.论文首先介绍了深度学习网络框架以及原理,并对随机生成的含典型病害铁路路基模型进行时域有限差分法正演,获得探地雷达仿真数据;然后将仿真数据预处理后导入深度学习网络中训练,并且利用测试集中数据验证该深度学习网络对一维探地雷达信号的噪声压制效果;最后,从图像上分析并评价了深度学习处理铁路路基雷达信号噪声压制的效果.采用上述方法处理后的雷达图像中异常体反射信号明显,同轴信号连续清晰,图像分辨率提高.研究结果表明该处理方法能有效压制轨枕强干扰信号,显著提高了雷达信号的信噪比,而且训练好的网络能实时处理海量的路基检测资料.Ground Penetrating Radar(GPR)with fast,noncontact,nondestructive characteristics,has become an important means of railway subgrade disease detection.However,the GPR signal is contaminated by electromagnetic reflection clutters from various sources in the railway subgrade detection.Especially,the strong reflection from the rail sleeper conceals the weak signal reflexed from the subgrade medium layer disease.In this paper,we have developed a method for suppressing the strong reflection clutters from rail sleepers based on deep learning.The paper firstly introduces the framework and principle of deep learning network.Next,the GPR simulation data to the random of railway subgrade model with typical defects is obtained by FiniteDifference TimeDomain(FDTD)forward method.Then the simulation data are preprocessed and imported into the deep learning network for training.And the noise suppression result for the onedimensional GPR signal is verified using the test set data by the deep learning network.Finally,the effect of deep learning processing on noise suppression of railway subgrade radar signals is analyzed and evaluated in terms of images.The signal reflected by anomalous body is obvious in the GPR image after processed by the above method.The coaxial signal is continuous and clear,and the image resolution is improved.The research results show the deep learning method can effectively suppress the strong interference reflected from the railway sleepers,the signaltonoise ratio of the GPR signal is significantly improved.Moreover,the trained network can process the massive amount of subgrade detection data in real time.

关 键 词：深度学习 探地雷达 铁路路基病害检测 轨枕噪声压制 

分 类 号：P631[天文地球—地质矿产勘探]