Dipole Models for UXO Discrimination at Live Sites - Pole Mountain

Abstract : The demonstration described in this report was conducted at the Pole Mountain, Wyoming, under project ESTCP MR-201159 Dipole Models for UXO discrimination at Live Sites. It was performed under the umbrella of the ESTCP Discrimination Study Program. The objective of the MR-201159 project is to demonstrate the application of feature extraction and statistical classification to the problem of UXO classification. At the Pole Mountain site, the objective was to discriminate targets of interest (TOI) (including 37 mm, 57 mm, 60 mm, 75 mm targets, a stokes mortar and a small industry standard object (ISO)) from non-hazardous shrapnel, range and cultural debris. In this report, we describe the performance of classification techniques that utilized data from a Geonics EM61 cart deployed in full coverage and MetalMapper data acquired in a cued interrogation mode. The classification techniques applied to the Pole Mountain data use features extracted from a dipole model fit to the observed data. Features used were the dipole polarizability tensor for the MetalMapper data and measures of the polarizability magnitude and decay rate for the Geonics EM61 data. From the extracted feature vectors the following prioritized dig-lists were created: (i) EM61 cart data ranked by polarizability time-decay and data amplitude; (ii) MetalMapper statistical classification using a Support Vector Machine applied to polarizabilities; and (iii) MetalMapper classification using a library matching method. All model fits and discrimination analysis were performed using the Sky classification software suite (UXOLab) that was jointly developed by UBC-GIF and Sky Research. The blind-test for the MetalMapper was split into two parts: Year 1 and Year 2. The Year 1 data set consisted of anomalies in the Southern portion of the site and the Year 2 data set consisted of anomalies in the Northern portion of the site. Year 1 targets were analyzed and scored first.