基于AV-GPR的路下空洞识别技术研究

城市道路塌陷灾害频发，阵列式车载探地雷达已成为城市道路病害普查检测的首选技术手段，但实际探测中，对地下空洞目标的检测和识别较为困难，依赖于人为判读和解译的方法效率非常低且常常导致漏检或虚检。本项目基于深度学习方法对AV-GPR采集的三维雷达回波数据进行分析识别，首先以城市道路地下各种空洞目标三维正演模拟为基础，结合大量实测数据提取的背景噪声合成更符合实际探测的三维空洞目标雷达回波图谱，作为深度学习的数据集；其次研究了利用鲁棒成分分析法（PRCA）和极值包络自动增益算法去除杂波干扰凸显目标信号；然后利用三维卷积神经网络对空洞目标进行分类识别，并利用数据集和实测空洞数据进行验证；最后研究在定期复测中利用深度学习算法进行对准比配，利用数据的相关性探测空洞缺陷。基于以上研究，实现路下空洞目标的自动识别，解决地下空洞探测中解译困难和判读不准确的问题，为城市道路安全这一重大民生需求提供技术支撑。

In view of current situation that urban road collapses frequently occur, the array vehicle ground penetrating radar (AV-GPR) is able to detect the subterranean cavity targets under urban roads quickly, efficiently and without mistake, becoming the first choice of technical means for the survey and detection urban road diseases. However, due to the complexity of the underground target’s location the detection can be easily affected by the environmental factors. Furthermore, the shapes of the subterranean voids are quite different from each other. Therefore, it is difficult to detect and identify the subterranean void target in the actual detection situation, and methods that rely on human interpretation and judgement are not very efficient and therefore often leads to false alarm or missing alarm. This project analyzes and identifies the 3D radar echo data collected by AV-GPR based on of the depth learning method. First, by utilizing the 3D forward simulation of various void targets under urban roads as the basis and combining it with the background noise extracted from a large number of measured data, the 3D radar echo map of void targets, which is more suitable for the actual detection, are synthesized and then used as the data set of depth learning. Next, the robust principal component analysis (RPCA) and the extremum envelope auto gain algorithm are used to remove the clutter and highlight the target signal. Then, the 3D convolution neural network is used to classify and identify the void target, also, the data set and the measured void data are used for verification. Finally, the depth learning algorithm is used for alignment and matching in the periodic retest, and the void defects can be found according to the changes of data. Based on the research above, we can realize the automatic recognition of the subterranean void target, solve the problem of interpretation difficulty and inaccurate judgement in the detection of subterranean void targets, and provide technical support for a major social demand—urban road safety.