Inversion of underground pipelines using multi-scale GAN

To improve the accuracy of underground pipelines detection in complex soil environments, this paper presents an end-to-end inversion method using generative adversarial networks (GAN) to convert the B-Scan data of ground penetrating radar (GPR) into the geometry and position of pipelines. In the Peplinski soil background, an electromagnetic simulation dataset for iron pipelines was constructed using GprMax. The accurate inversion of underground pipelines’ geometry and position was achieved through the training of the Pix2PixHD network model. The experimental results show that the pipeline coordinate inversion error of this method is controlled within 3% on the test set, the contour reconstruction matches the real model, the peak signal to noise ratio (PSNR) is 40.8dB and the structural similarity index measurement (SSIM) reaches 99.53%, verifying the similarity between the generated image and the original model. This method avoids limitations of traditional inversion and provides a high-precision, high-efficiency technical solution for the safe management of urban underground pipelines.