Subsurface Diagnosis With Time-Lapse GPR Slices and Change Detection Algorithms

Abstract

This article explores the capability of applying timelapse ground penetrating radar (GPR) data to investigate the health condition of an urban subsurface. A workflow is proposed to semi-automatically extract changes from time-lapse GPR C-scans. The developed workflow consists of two main steps, in which the first step is image registration and intensity normalization. The workflow uses benchmark points on the ground to normalize the global intensity of time-lapse GPR C-scans. The second step classifies pixels into change or unchanged group. Two kinds of information are considered to construct two difference-maps: changes in the image intensity and the object structure. K-means clustering is responsible for extracting pixels that possess both intensity changes and object structure changes-where potential subsurface defects most likely occurred. The workflow was verified by a site experiment, and the area of excavation with pipe replacement was successfully identified. The performance of the proposed workflow was promising in excluding small and random scattering noise, which was the main challenge in a time-lapse GPR survey. The article serves as a prototype and demonstrates the feasibility and necessity of conducting temporal diagnosis on the subsurface structure. Index Terms-Ground penetrating radar (GPR), subsurface diagnosis, temporal change detection, time-lapse. I. INTRODUCTION M ODERN cities are facilitated by a large number of underground utilities. However, the management of these subsurface infrastructures is complicated, and the work of managing invisible underground utilities has proven to be especially demanding and costly. Without proper diagnosis and maintenance, ageing utilities can suffer from various modes of failures, bringing urban hazards such as land subsidence, the collapse of infrastructure, and flooding. They can cause not only financial loss, but also causalities. Conducting regular health checks for

Efficient Denoising of Multidimensional GPR Data Based on Fast Dictionary Learning

“…precision and high resolution and is often used in environmental, geological applications [1][2][3] and engineering exploration [4][5][6], and other fields. However, in field acquisition, due to the complex distribution of underground media, environmental interference, and other factors, the GPR data often contain strong background clutter and various random noises [7], [8].…”