Research on Clutter Suppression of GPR Data in Tunnel Karst Based on Deep Learning
To enhance the accuracy of the GPR(Ground-Penetrating Radar)data in karst regions and reduce the impact of clutter on GPR data,thereby precisely delineating the target anomalies,this paper presents a data processing method based on VAE-RFDB-UNet deep learning.This approach begins with the use of a VAE(Variational Autoencoder)to extract key information from the GPR data,achieving preliminary suppression of clutter.Subsequently,UNet network enhanced by an RFDB module performs deep feature learning on the VAE-processed data,forming a cascaded network that further suppresses clutter interference and enhances target signal response.The final output is a reconstructed GPR data with improved SNR(Signal-to-Noise Ratio).The research results show that:1)The effectiveness and feasibility of the algorithm are demonstrated using both numerical and field GPR data.2)Results from numerical GPR data show that the karst reflections maintain a complete and continuous hyperbolic shape with significant clutter suppression by post-processing.3)Analysis of field GPR data indicates that the algorithm effectively processes field data,allowing accurate inference of karst development ahead of the tunnel face based on the processed profiles.This approach offers a novel perspective for processing GPR data in karst regions.
NETL
NSTL
万方数据
