Underground Pipeline Precise Positioning Method Based on GPR Bscan Images

Ground-penetrating radar (GPR) enables the positioning of underground pipelines without excavation. However, the accuracy and errors of existing methods are still insufficient due to the influence of strong clutter in actual environments. In this paper, an underground pipeline precise positioning method based on GPR Bscan images is proposed. First, the YOLOv8 neural network is employed for initially pipeline positioning. Next, by combining network post-processing with curve fitting, the pipeline depth is estimated with high accuracy, enabling precise pipeline positioning. Finally, in validation using measured data, the proposed method achieved a mean average precision (MAP) of 96.8% in the initially pipeline positioning step and a relative depth estimation error of 4.84% in the precise positioning step.