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

Forum topic · 小凯 · 2026-08-21

Summary

Ground Penetrating Radar (GPR) is a widely used non-destructive sensing technology for subsurface inspection in civil and transportation engineering, but its large-scale use in automated pavement inspection faces two challenges: scarcity of annotated real-world datasets and the lack of deep learning models tailored to 3D GPR data. This arXiv paper (2608.19177) by Pan et al. addresses both limitations. First, it introduces a cost-effective data preparation pipeline that integrates orthomosaic RGB imagery with 3D GPR scans, transferring labels of surface-visible defects from pavement images to aligned GPR segments, enabling efficient large-scale annotation of data collected on an operating highway. Second, it proposes a dedicated 3D convolutional neural network (CNN) architecture combining residual connections, hybrid kernel sizes, and channel and depth attention mechanisms for enhanced feature representation. Evaluated on a binary classification task detecting patches and cracks in pavement structures, the proposed network outperforms baseline architectures across multiple metrics, and ablation studies confirm the effectiveness of each architectural component. The work contributes both a scalable real-world dataset generation method and a novel deep learning framework for GPR-based pavement defect recognition.

Paper Overview

Field: Computer Vision (CV)
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

Background

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, large-scale automated GPR inspection faces two key challenges:

1. The scarcity of annotated real-world datasets. 2. The lack of deep learning models designed for the unique characteristics of 3D GPR data.

Contributions

Cost-effective data preparation pipeline — The study integrates orthomosaic RGB imagery with 3D GPR scans to generate annotated 3D GPR datasets. Using aligned segments of RGB and GPR data, labels of surface-visible defects are transferred from pavement surface images to the corresponding GPR segments, enabling efficient large-scale annotation of real-world data collected on an operating highway.

Novel 3D CNN architecture — A dedicated 3D convolutional neural network is proposed, combining:

Residual connections
Hybrid convolution kernel sizes
Channel and depth attention mechanisms

These components enhance feature representation and defect classification.

Results

The model was evaluated on a binary classification task detecting patch and crack defects in pavement structures. Experimental results show the proposed network outperforms baseline architectures across multiple evaluation metrics, and ablation studies confirm the effectiveness of the designed architectural components.

Conclusion

The work contributes a scalable and practical method for generating real-world annotated 3D GPR datasets, along with a novel deep learning framework for automated pavement defect recognition.

---

*Auto-collected on 2026-08-21*