Abstract:The 3D ground penetrating radar (GPR) is the main method for detecting underground cavities in urban roads. Due to the weak reflected signal energy of deep road cavities with depths exceeding 2.5 m, there is a significant shortage of available training samples. Existing identification algorithms primarily focus on the detection of shallow road cavities. As a result, the accuracy of deep cavity identification using 3D GPR is low, and there is a lack of effective intelligent algorithms for deep cavity identification. To address these challenges, this study integrates the smooth texture features and the abundant amplitude and phase spectrum features inherent in deep cavity GPR signals. Utilizing the time-frequency features of radar signals, this study has proposed an intelligent identification algorithm for the deep road cavity based on a Multi-Channel and Dimensional Time-Frequency Convolution Neural Network (MCD-TF CNN). Firstly, using MCD-FT CNN as the cavity value discriminator, inverse reinforcement learning is performed on the cavity region to obtain the value evaluation method of the 3D GPR cavity region. Subsequently, the discriminator is applied to deep radar data for cavity detection and interacts with the value discriminator through 3D target region range adjustment actions. The interaction through region adjustments aims to maximize the value of cavity areas within the discriminator relative to the target detection area. It can ensure the inclusion of cavity regions in the detection results, thereby achieving the goal of intelligent deep cavity recognition. This algorithm employs the thought of global relative optimality from reinforcement learning, addressing the limitations of existing algorithms that rely on extracting absolute features and thereby enhancing the accuracy of deep cavity identification.

Development of a Background Filtering Algorithm to Improve the Accuracy of Determining Underground Cavities Using Multi-Channel Ground-Penetrating Radar and Deep Learning

Lightweight Deep Learning Model for Identifying Tunnel Lining Defects Based on GPR Data

Deep l earning-based a utonomous u nderground c avity d etection using 3 D GPR

Deep learning-based automated underground cavity detection using three-dimensional ground penetrating radar

A novel 3D GPR image arrangement for deep learning-based underground object classification

Triplanar Imaging of 3-D GPR Data for Deep-Learning-Based Underground Object Detection

Research on 3D ground penetrating radar deep underground cavity identification algorithm in urban roads using multi-dimensional time-frequency features

3D GPR Image-based UcNet for Enhancing Underground Cavity Detectability

Defects identification and location of underground space for ground penetrating radar based on deep learning

FM-GAN: Forward Modeling Network for Generating Approaching-Reality B-Scans of GPR Pipelines With Transformer Tuning

3D ground penetrating radar cavity identification algorithm for urban roads using transfer learning

An Encoding-based Back Projection Algorithm for Underground Holes Detection via Ground Penetrating Radar

RCE-GAN: A Rebar Clutter Elimination Network to Improve Tunnel Lining Void Detection from GPR Images

Underground Object Classification Using Deep 3-D Convolutional Networks and Multiple Mirror Encoding for GPR Data

Underground Object Classification for Urban Roads Using Instantaneous Phase Analysis of Ground-Penetrating Radar (GPR) Data

3D ground penetrating radar road underground target identification algorithm using time-frequency statistical features of data

Automatic Defect Detection in Operational High-Speed Railway Tunnels Guided by Train-Mounted Ground Penetrating Radar Data

Deep Learning-Based Underground Object Detection for Urban Road Pavement

Underground Anomaly Detection in GPR Data by Learning in the C3 Model Space

Three-Dimensional Reconstruction of Retaining Structure Defects from Crosshole Ground Penetrating Radar Data Using a Generative Adversarial Network

Underground Mapping and Localization Based on Ground-Penetrating Radar