基于深度学习的隧道岩溶GPR数据杂波抑制研究

Research on Clutter Suppression of GPR Data in Tunnel Karst Based on Deep Learning

为提高岩溶地区GPR数据的解析精度,减少杂波对GPR剖面的影响,从而更准确地圈定目标异常,提出了基于VAE-RFDB-UNet深度学习的数据处理方法。该方法先通过VAE对GPR数据进行有效信号特征提取,实现杂波的初步抑制,后利用加载RFDB模块的UNet网络对VAE处理后的数据进行深度特征学习,形成联级网络,进一步抑制杂波干扰,增强目标信号响应,提高数据的信噪比。研究结果表明:1)使用模拟数据和实测数据证明了算法的有效性和可行性;2)数值模拟的处理结果显示,处理后的岩溶反射波双曲线形态完整、连续,杂波得到了明显抑制;3)实测数据的处理结果显示,该算法有效突出了目标异常信号,能依据处理后的剖面准确推断掌子面前方的岩溶发育情况。结论是提出的VAE-RFDB-UNet深度学习算法可为岩溶地区GPR数据处理提供一种新思路。

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.