An Adaptive Clutter-Immune Method for Pipeline Detection With GPR

The detection and localization of subsurface pipelines using ground-penetrating radar (GPR) are challenging endeavors primarily impeded by the formidable clutter interference. Therefore, the effective removal of background clutter and the enhancement of target response are essential. This study proposed an efficient clutter removal neural network called Clutter-immune Net (CI-Net). CI-Net incorporated a convolutional triplet attention module into the residual module (RSU) to identify critical regions with different scale features. The network's adaptive mechanism fused the outputs of multiscale RSU modules to obtain more accurate and comprehensive target responses. A combination of smooth L1 loss (SLL) and multiscale structural similarity loss (MS-SSIM) was adopted to improve the network's optimization capability, which has proven effective. Furthermore, to improve the generalization of clutter processing, target components of simulated data are integrated into the target-free real data during the acquisition of the dataset. The proposed method is compared with existing methods using simulated and measured data. The peak signal-to-noise ratio (PSNR), structural similarity (SSIM), image entropy, and visual quality results confirm that the method exhibits better clutter-immune performance. The performance of the proposed method is assessed for underground pipelines in different scenes, and the results showed that the model could effectively remove clutter in various environmental settings. This study lays the foundation for further research on GPR inversion tasks.