ABSTRACT

Missing information in ground-penetrating radar (GPR) data is a commonly encountered phenomenon during field measurements and it highly affects the subsequent processing steps of the GPR data, such as migration, clutter removal, target identification, etc. Since the GPR field tests are time consuming and hard to repeat, it is better to use the measured data even if they contain partial missing information. Due to the common problems in GPR field measurements, the missing information can occur in both pixel and column-wise and there are many methods proposed to solve them. In this study, we selected the best matrix completion methods for GPR data which provide satisfactory results. Some of the methods are already applied to GPR problem however there are no extensive comparisons available and recently proposed ones are for the first time used, which are the novelty of this study. Among these methods, nuclear norm minimization (NNM) and non-negative matrix completion (NMC) outperform others for the pixel-wise and the column-wise cases. Both simulated and real dataset results show that for moderate missing information cases NNM can be selected however for extreme cases NMC gives better results.

摘要

探地雷达（GPR）数据中的信息丢失是在野外测量过程中经常遇到的现象，并且会严重影响GPR数据的后续处理步骤，例如迁移，杂波去除，目标识别等。既耗时又难以重复，即使测量数据中包含部分缺失信息，也最好使用测量数据。由于GPR现场测量中的常见问题，缺少的信息可能会在像素方向和列方向出现，并且提出了许多解决方法。在这项研究中，我们为GPR数据选择了最佳的矩阵完成方法，这些方法可提供令人满意的结果。有些方法已经应用于GPR问题，但是目前尚无广泛的比较方法，并且最近首次提出了比较方法，这是这项研究的新颖之处。在这些方法中，在像素级和列级情况下，核规范最小化（NNM）和非负矩阵完成（NMC）优于其他方法。模拟数据集和实际数据集结果均表明，对于中等程度的信息丢失情况，可以选择NNM，但是对于极端情况，NMC可以提供更好的结果。