Share Email Print

Proceedings Paper

Use of EMI response coefficients from spheroidal excitation and scattering modes to classify objects via SVM
Author(s): Beijia Zhang; Kevin O'Neill; Tomasz M. Grzegorczyk; Jin Au Kong
Format Member Price Non-Member Price
PDF $14.40 $18.00
cover GOOD NEWS! Your organization subscribes to the SPIE Digital Library. You may be able to download this paper for free. Check Access

Paper Abstract

Electromagnetic induction (EMI) has been shown to be a promising technique for unexploded ordnance (UXO) detection and discrimination. The excitation and response of a UXO or any other object to EMI sensors can be described in terms of scalar spheroidal modes consisting of associated Legendre functions. The spheroidal response coefficients Bjk correspond to the kth spheroidal response to the jth spheroidal excitation. The Bjk have been shown to be unique properties of an object, in that objects producing different scattered fields must be characterized by different Bjk. Therefore, the Bjk coefficients may be useful in discrimination. We use these coefficients rather than dipole moments because they are part of a physically complete, rigorous model of the object's response. Prolate spheroidal coordinate systems recommend themselves because they conform most readily to the proportions of objects of interest. In clearing terrain contaminated by UXO, the ability to distinguish larger buried metallic objects from smaller ones is essential. Here, a Support Vector Machine (SVM) is trained to sort objects into different size classes, based on the Bjk. The classified objects include homogeneous spheroids and composite metallic assemblages. Training a SVM requires many cases. Therefore, an analytical model is used to generate the necessary data. In simulation studies, the SVM is very successful in classifying independent sets of objects of the same type as the training set. Furthermore, we see that the Bjk are not related to size or signal strength of the object in any simple or visually discernible way. However, SVM is still able to sort the objects correctly. Ultimately, the success of the SVM trained with synthetic (model derived) data will be evaluated in application to data from a limited population of real objects, including UXO.

Paper Details

Date Published: 16 May 2006
PDF: 11 pages
Proc. SPIE 6217, Detection and Remediation Technologies for Mines and Minelike Targets XI, 621708 (16 May 2006); doi: 10.1117/12.668122
Show Author Affiliations
Beijia Zhang, Massachusetts Institute of Technology (United States)
Kevin O'Neill, Dartmouth College (United States)
Engineer Research and Development Ctr. (United States)
Tomasz M. Grzegorczyk, Massachusetts Institute of Technology (United States)
Jin Au Kong, Massachusetts Institute of Technology (United States)

Published in SPIE Proceedings Vol. 6217:
Detection and Remediation Technologies for Mines and Minelike Targets XI
J. Thomas Broach; Russell S. Harmon; John H. Holloway, Editor(s)

© SPIE. Terms of Use
Back to Top