Study on Pipe Diameter Identification Based on A-Scan Information on GPR Hyperbola

Abstract

Accurate diameter identification of buried pipelines remains challenging in complex municipal environments because ground-penetrating radar (GPR) echoes are often weak and hyperbolic reflections are frequently incomplete. To address this problem, this study proposes a diameter identification method that integrates five-point hyperbolic A-scan features with burial depth information. Faster R-CNN is used to locate the hyperbolic region and remove redundant background information. HRNet is then employed to detect five keypoints, including the hyperbola vertex, the left and right diffraction midpoints, and the left and right tail points, from which the corresponding A-scan signals are extracted as lightweight features. The burial depth of the pipeline is subsequently estimated using LSTM and introduced as prior information. Finally, a Curve-Net regression network fuses the five key A-scans and the burial depth feature for diameter regression. Experimental results show that the proposed method achieves an MAE of 0.0159 m, an RMSE of 0.0202 m, and an R2 of 0.8354 on the test set. In field validation, the maximum relative error is 8.50%, with an absolute error of only 1.7 cm. These results demonstrate that the proposed method can effectively improve the accuracy of concealed pipeline diameter identification in small and medium-sized, shallow-buried municipal pipeline scenarios, and exhibits promising engineering application value.