Signal processing and fusion study for subsurface object detection using ground penetrating radar

Abstract

[ACCESS RESTRICTED TO THE UNIVERSITY OF MISSOURI AT AUTHOR'S REQUEST.] The detection of subsurface objects with ground penetrating radar (GPR) continues to be a difficult problem partially due to variation of their content and depths of burial. Three detection techniques have been developed with the goal of building detectors robust to these issues. Two of these techniques are based of autoregressive (AR) modeling and are designed to detect specifically shallow buried low-metal content objects. The third technique utilizes non-negative matrix factorization (NMF) to generate a representation of the component parts of the GPR signal. The NMF technique can detect both low and non-metal objects over a wide depth range. In addition to the detection techniques, a study on detector fusion is provided which combines results of multiple detectors including AR and NMF. This study has resulted in fusion algorithms that can increase the detection performance beyond that of the NMF detector.