近年来，探地雷达（Ground-Penetrating Radar , GPR）可以作为一种非侵入式地探测铁路路基的方法在路基病害的研究中变得越来越重要。路基病害的探地雷达剖面图像包含复杂的双曲线信号，会由于病害的大小，埋深，介质背景的影响发生信号变形。由于多种变形因素影响，通过目视解译判别和分析路基病害类型、位置以及病害程度较为困难严重影响工作效率及成本。因此，如何快速、精确地从GPR图像中识别铁路路基的病害成为一项挑战。本文首先提出利用专业的探地雷达数值模拟软件GprMax来模拟典型路基病害——翻浆冒泥的情景，通过控制翻浆冒泥病害特征的各种参数模拟出3000张探地雷达剖面图像，随后利用生成的模拟数据对基于深度学习的目标检测模型，即YOLOv5（You Only Look Once）模型进行训练，将训练好的模型用于自动快速识别GPR探测路基剖面中的翻浆冒泥病害。并在实测数据上进行评估，评估结果显示精度和召回率分别高达 96.3%和 98.8%，在一定的置信度内基本实现对实测数据病害的识别，实现了较好的识别效果，也为现有对铁路路基病害的检测研究提供了一定的思路。

In recent years, Ground-Penetrating Radar (GPR) has become increasingly important as a non-intrusive method for detecting railway trackbed conditions in the study of trackbed diseases. GPR profile images of trackbed diseases contain complex hyperbolic signals, which undergo signal deformation due to factors such as the size of the disease, burial depth, and medium background. Due to the influence of multiple deformation factors, visually interpreting and analyzing the types, locations, and extents of trackbed diseases is challenging, severely impacting work efficiency and costs. Therefore, how to rapidly and accurately identify trackbed diseases from GPR images has become a challenge. This paper first proposes to use professional GPR numerical simulation software, GprMax, to simulate typical trackbed diseases—such as mud heaving and mud pumping scenarios—by controlling various parameters of the mud heaving and mud pumping disease characteristics to simulate 3000 GPR profile images. Subsequently, generated simulation data is utilized to train a deep learning-based object detection model, namely YOLOv5 (You Only Look Once) model. The trained model is then used to automatically and rapidly identify mud heaving and mud pumping diseases in GPR-detected trackbed profiles. Evaluation on real measured data shows precision and recall rates of 96.3% and 98.8%, respectively, achieving basic identification of measured data diseases within a certain confidence level. This approach demonstrates good identification performance and provides insights for existing research on detecting railway trackbed diseases.