Accounting for Hydrologic State in Ground-Penetrating Radar Classification Systems
Abstract
The objectives of this work were to: (1) evaluate the influence of hydrologic processes (i.e., changes in soil water content) on ground-penetrating radar (GPR) signals, particularly those associated with landmines, and (2) investigate the potential for developing contextual GPR classification systems by accounting for changes in environmental state (i.e., soil water content) using hydrologic modeling. The estimation of soil water content is a major focus of this work since this property is closely related to EM wave propagation in soils (i.e., dielectric constant, electrical conductivity, wave velocity), which control radar responses. The general research hypothesis guiding this work is that accounting for hydrologic state in classification systems will allow for improved generalization of landmine classification tools to a broader range of sites under varying operational conditions. The focus of the research was on two-dimensional imaging and simulation, though we also demonstrated the value of three-dimensional GPR imaging for improved object detection and characterization of flow processes in soils and around buried objects. Overall we found that accurate estimates of water content can be derived from GPR signals and that accounting for the water content of a soil within a contextual classification system is likely to improve classification results. The classification gains observed in this study were somewhat modest when comparing a contextual classifier to a non-contextual classifier that was trained over targets observed for a large set of hydrologic conditions. Both of these approaches significantly outperformed a classification strategy that first attempted to correct GPR signals observed at arbitrary conditions to a single hydrologic reference state. We are continuing to evaluate the significance of our results to scenarios representative of a broader range of conditions than those considered in this study.
Document Details
- Document Type
- Technical Report
- Publication Date
- Apr 22, 2014
- Accession Number
- ADA612641
Entities
People
- Stephen Moysey
Organizations
- Clemson University