Ground Penetrating Radar Signal Denoising Based on Multi-Scale Feature Fusion Network

With the rapid development of electronic information technology, Ground Penetrating Radar (GPR) technology has become an indispensable tool in fields such as urban construction, archaeology, and the military. However, due to the heterogeneity of underground media, scattering of electromagnetic waves, and equipment errors, GPR signals are often disturbed by noise, affecting detection accuracy. Therefore, this paper designs a multi-scale feature fusion network to enhance the robustness of denoising models for denoising GPR signals with different characteristics. The network is constructed based on the Inception architecture and consists of three parts: an encoder structure for extracting multi-scale features of GPR signals, a disentangler structure for classifying data features and generating corresponding feature masks, and a decoder for reconstructing the denoised signal. Additionally, ID-Net can also extract noise features, enhancing the model's interpretability. Experiments have proven that ID-Net outperforms traditional denoising models in removing GPR noise and demonstrates good denoising capabilities on datasets with different characteristics, proving the robustness of ID-Net in denoising GPR noise signals with different features.