Discriminating munitions of all sizes is a stated need for the Munitions Response Program. To address this need, a series of live site demonstrations has been initiated by ESTCP to facilitate the adoption of classification technologies and to document classification performance at sites of varying ordnance and clutter distributions, as well as different site environments. The pace of activities and the volume of data being collected increases dramatically with each passing demonstration. Additionally, the number of firms actively participating in the classification studies is also increasing steadily, as more and more commercial contracting firms are naturally participating. This increased pace and activity level amplifies the prospect of missing lessons learned or not taking the time required to systematically delve into classification challenges and failures.

The objective of this project is to investigate and understand the nature and extent of classification challenges and failures as revealed in data collected in support of ESTCP's live site demonstrations and to demonstrate the methods developed during a subsequent live site demonstration.

Recent classification demonstrations have shown that new multi-axis EMI sensor systems can be used to successfully distinguish between buried munitions and clutter when they are operated in a cued interrogation mode most of the time. Classification errors result, however, if the measured data have unrecognized problems, if non-robust analysis methods are used, if improper training data (labels) are used, or if the data are not leveled correctly (viz., removing background and system response).

The technologies involved in this effort include the latest multi-axis electromagnetic induction (EMI) sensor systems and state of the art data analysis algorithms. These technologies have shown great promise and potential with regard to classifying a visually obscured piece of metal as unexploded ordnance (UXO) or not. When deployed in a cued interrogation, the EMI sensor is positioned over an anomaly of interest, the transmit coils are excited sequentially, and the response from the target at the various receive coils is recorded. These data are inverted to estimate source object parameters, which are then supplied to classification engines that decide whether they are more likely associated with munitions or clutter. The characterization phase is accomplished using physics-based solvers that invert the spatially-registered EMI data to determine the target's location, orientation, and principal axis polarizabilities. The inversion is based on a dipole response model in which the polarizabilities specify the target's intrinsic electromagnetic response.

By capturing and documenting lessons learned during ESTCP's live site demonstrations, the project team will improve classification performance, advance collective understanding, and further technology transfer to the user community. (Anticipated Project Completion - 2014)