Approach to Explosive Hazard Detection Using Sensor Fusion and Multiple Kernel Learning with Downward-Looking GPR and EMI Sensor Data Anthony Pinar a , Matthew Masarik b , Timothy C. Havens a,c , Joseph Burns b , Brian Thelen b , and John Becker a a Department of Electrical and Computer Engineering, Michigan Technological University, Houghton, MI USA b Michigan Tech Research Institute, Michigan Technological University, Ann Arbor, MI USA c Department of Computer Science, Michigan Technological University, Houghton, MI USA ABSTRACT This paper explores the effectiveness of an anomaly detection algorithm for downward-looking ground penetrating radar (GPR) and electromagnetic inductance (EMI) data. Threat detection with GPR is challenged by high responses to non-target/clutter objects, leading to a large number of false alarms (FAs), and since the responses of target and clutter signatures are so similar, classifier design is not trivial. We suggest a method based on a Run Packing (RP) algorithm to fuse GPR and EMI data into a composite confidence map to improve detection as measured by the area-under-ROC (NAUC) metric. We examine the value of a multiple kernel learning (MKL) support vector machine (SVM) classifier using image features such as histogram of oriented gradients (HOG), local binary patterns (LBP), and local statistics. Experimental results on government furnished data show that use of our proposed fusion and classification methods improves the NAUC when compared with the results from individual sensors and a single kernel SVM classifier. Keywords: explosive hazards, ground penetrating radar, electromagnetic inductance, unsupervised learning, signal processing, sensor fusion 1. INTRODUCTION Buried explosive hazards represent one of the greatest threats to human life in modern combat as well as a danger to civilian populations residing in former war zones. Every month, explosive devices kill an average of 310 people and wound 833 others. 1 Several systems have been investigated as a means of detecting these hazards. Among them are ground-penetrating-radar (GPR), infrared (IR) and visible-spectrum cameras, and acoustic and seismic technologies. 2–4 While forward-looking GPR systems offer standoff distances between the radar and targets, downward-looking radar is more compact and can be adopted to autonomous platforms to remove the need for human presence. Downward-looking radar also has the added benefit of receiving much more of the transmitted radar energy than forward-looking systems due to geometry; most of the energy emitted from forward-looking systems reflects off the ground away from the receiver. Additionally, the downward-looking GPR data collection systems can employ multiple sensors simultaneously, allowing sensor fusion methods to be applied to the collected data. We describe our image formation process in Section 2. The constant false alarm rate (CFAR) prescreener we employ is described in Section 3, and in Section 4 we describe an established sensor fusion algorithm known as run packing to combine the detection results from multiple sensors. We also explore the use of composite confidence maps to blend the detections from multiple sensors into a single hit list. Section 5 describes the context-based features we extract from each candidate hit location, where each feature is extracted from both the hit location itself and the region surrounding it. The results of our experiments are given in Section 7, followed by the conclusion in Section 8. Further author information: (Send correspondence to T. C. Havens) T.C. Havens: E-mail: thavens@mtu.edu, Telephone: 1 906 487 3115 A. Pinar: E-mail: ajpinar@mtu.edu DISTRIBUTION STATEMENT A: Approved for public release: distribution unlimited. Detection and Sensing of Mines, Explosive Objects, and Obscured Targets XX, edited by Steven S. Bishop, Jason C. Isaacs, Proc. of SPIE Vol. 9454, 94540B © 2015 SPIE · CCC code: 0277-786X/15/$18 · doi: 10.1117/12.2176856 Proc. of SPIE Vol. 9454 94540B-1 Downloaded From: http://proceedings.spiedigitallibrary.org/ on 06/04/2015 Terms of Use: http://spiedl.org/terms