Ground penetrating radar (GPR) is a geophysical exploration technique used in a wide range of applications such as polar exploration, hydrogeological surveys, and archeological prospection. In recent years, it has been applied to tunnel construction for forecasting geological anomalies in front of the tunnel face. However, the application of GPR, forecasting geological anomalies in karst tunnels, has always been restricted by the problem of ambiguity. In the paper, based on the GPR investigation of highway tunnels in the karst areas in Guangxi, China, several GPR data of typical karst geological anomalies were selected as the analysis objective. The purpose of this study is twofold: to analyze and summarize the GPR attribute characteristics of typical geological anomalies by studies of laboratory tests and to introduce a new method for identifying different types of karst geological anomalies automatically in karst tunnel construction. Firstly, the attribute analysis technology of GPR is introduced, and the propagation law of electromagnetic waves in karst geological anomalies is studied from three perspectives: the time domain, frequency domain, and time-frequency domain by using physical laboratory simulations. Then, an intelligent identification model for typical karst geological anomalies is established in accordance with a Gaussian process (GP). Results from the Gaussian classification of field cases indicate that the type of typical karst geological anomalies can be effectively identified via comparing model prediction probability. The findings of this research allow for quantitative description of the GPR interpretation of typical karst geological anomalies in tunnel advanced prediction.

中文翻译：

---

基于GPR属性和高斯过程的岩溶隧道施工典型异常识别方法

探地雷达（GPR）是一种地球物理勘探技术，广泛用于极地勘探，水文地质调查和考古勘探等领域。近年来，它已被应用到隧道建设中以预测隧道工作面前的地质异常。但是，GPR在预测喀斯特隧道地质异常方面的应用一直受到歧义问题的限制。本文基于广西喀斯特地区公路隧道的GPR调查，选取了典型喀斯特地质异常的GPR数据作为分析目标。这项研究的目的是双重的：通过实验室试验研究，分析和总结了典型地质异常的GPR属性特征，为岩溶隧道施工中自动识别不同类型的岩溶地质异常提供了一种新方法。首先介绍了GPR的属性分析技术，并通过物理实验室仿真从时域，频域和时频域三个角度研究了电磁波在岩溶地质异常中的传播规律。然后，根据高斯过程（GP）建立了典型的喀斯特地质异常的智能识别模型。高斯现场案例分类的结果表明，通过比较模型预测概率可以有效地识别典型的喀斯特地质异常类型。这项研究的发现可以定量描述隧道超前预报中典型喀斯特地质异常的GPR解释。