Image-Guided Pavement Defect Recognition in GPR Data with a Novel 3D Deep Learning Model

Forum topic · 小凯 · 2026-08-21

Summary

This paper addresses two key barriers to large-scale automated pavement inspection using Ground Penetrating Radar (GPR): the scarcity of annotated real-world datasets and the lack of deep learning models tailored to 3D GPR data. The authors introduce a cost-effective data preparation pipeline that integrates orthomosaic RGB imagery with 3D GPR scans, using pavement surface images as a reference to transfer labels of surface-visible defects to corresponding aligned GPR segments—enabling efficient large-scale annotation on a real-world dataset collected from an operating highway. They further propose a dedicated 3D convolutional neural network combining residual connections, hybrid kernel sizes, and depth and channel attention mechanisms for enhanced feature representation and defect classification. The model was evaluated on a binary classification task detecting patches and cracks in pavement structures, outperforming baseline architectures across multiple metrics, with ablation studies confirming the effectiveness of each architectural component. arXiv: 2608.19177.

Paper Overview

Field: Computer Vision Authors: Yuandong Pan, Linjun Lu, Mudan Wang, Florian Noichl, Fan Xue, Brian Sheil, Lavindra de Silva, André Borrmann, Ioannis Brilakis Published: 2026-08-19 arXiv: 2608.19177

Abstract

Ground Penetrating Radar (GPR) is a widely adopted non-destructive sensing technology for subsurface inspection in civil and transportation engineering. Despite its potential for pavement condition assessment, the large-scale application of GPR in automated inspection has two key challenges: the scarcity of annotated real-world datasets and the lack of deep learning models designed for the unique characteristics of 3-Dimensional (3D) GPR data.

This study addresses these limitations by firstly introducing a cost-effective data preparation pipeline that integrates orthomosaic Red Green Blue (RGB) imagery with 3D GPR scans to generate annotated 3D GPR datasets. The proposed method uses aligned segments of RGB and GPR data, taking pavement surface images as a reference to transfer labels of surface-visible defects to the corresponding GPR segments, enabling efficient large-scale annotation in a real-world dataset collected on an operating highway section.

In addition to the dataset contribution, the authors propose a specialized 3D convolutional neural network (CNN) architecture that combines residual connections, hybrid convolution kernel sizes, and depth and channel attention mechanisms to enhance feature representation and defect classification. The model was evaluated on a binary classification task detecting patch and crack defects in pavement structures. Experimental results show that the proposed network outperforms baseline architectures across multiple evaluation metrics, and ablation studies further confirm the effectiveness of the designed architectural components. This work contributes both a scalable and practical method for real-world dataset generation and a novel deep learning framework.

Key Contributions

1. Data pipeline: A cost-effective workflow that fuses orthomosaic RGB imagery with 3D GPR scans to produce annotated 3D GPR datasets from operating highway sections. 2. Label transfer: Surface images serve as a reference to transfer defect labels to aligned GPR segments, enabling large-scale annotation. 3. 3D CNN architecture: Residual connections, mixed kernel sizes, and depth/channel attention for improved defect classification (patches vs. cracks). 4. Validation: Outperforms baselines on multiple metrics; ablation studies verify each component's contribution.