关键词: 关键词: 探地雷达（GPR）, 杂波抑制, 细节增强网络, 特征-像素双级融合损失, 机场跑道

Abstract: To address the problem of complex background clutter and inter-layer strong reflections in airport runway ground-penetrating radar (GPR) data, an improved U-Net-based deep learning method for clutter suppression was proposed. A detail enhancement module (DE-Conv) was introduced at the skip connections of U-Net to improve multi-scale feature extraction, and a feature-pixel dual-level fusion loss function was designed using paired clutter-contaminated and clutter-free images to optimize training. High-dimensional features were extracted by a shared-weight encoder for feature-level loss computation, while pixel-level loss was calculated between decoder outputs and clutter-free simulation images. The experimental results show that the proposed method achieves a peak signal-to-noise ratio (PSNR) of 37.1147 dB and a structural similarity index (SSIM) of 0.9998 on synthetic data, and an average signal-to-clutter ratio (SCR) of 8.28 dB and improvement factor (IF) of 5.90 dB on real runway data, outperforming baseline models by 0.9528 dB, 0.0004, 6.58 dB, and 5.32 dB in these four metrics. Compared with RNMF, RPCA, and the deep learning-based CR-Net method, superior clutter suppression performance and computational efficiency are demonstrated. The effectiveness of the DE-Conv module and dual-level loss function is further validated through ablation experiments.

Key words: Keywords: ground penetrating radar (GPR), clutter suppression, detail enhancement network, dual-level fusion loss of feature and pixel, airport runway