ALLTEM UXO Discrmination Results from the Aberdeen Proving Ground Using a Hybrid Generalized Neural Analysis and Standard Dipole Inversion and Classification Scheme
Ted Asch, Michael James Friedel · 2011
An advanced multi-axis electromagnetic induction system, ALLTEM, has been specifically designed for detection and discrimination of unexploded ordnance (UXO). This work has been funded by ESTCP (Project MM-0809). ALLTEM uses a continuous triangle-wave excitation that measures the target step response rather than the more common impulse response. Ferrous and non-ferrous metal objects have distinct characteristic responses. The system multiplexes through all three orthogonal (Hx, Hy, and Hz axes) transmitting loops and records a total of 19 different transmitting (Tx) and receiving (Rx) loop combinations with a spatial data sampling interval of 20 cm. This paper presents some of the results of a demonstration and validation survey at the Aberdeen Proving Ground in March 2010. The U.S. Geological Survey operated ALLTEM with a Leica 1200 GPS over the Army's UXO Calibration and Blind Test Grids and the Direct Fire and Indirect Fire areas. Custom data analysis is conducted from within Oasis Montaj including importing survey data, gridding, noise analysis for threshold determination, automatic selection of targets, and inversion and classification. Batch inversion of selected targets with a prolate spheroid starting model followed by the application of neural SOM (self organizing map) algorithms is used to automatically classify the objects into targets of interest and those of clutter. The SOM process is unsupervised and completely data driven. The goal of learning in the SOM is to cause different parts of the network (the data) to respond similarly to certain input patterns. The result of this process is clustering of similar target data in different parts of the ‘map’. This SOM analysis, used in conjunction with the standard numerical dipole inversion results, provides a better understanding of what data is being provided to the SOM and to the numerical inversion with the result being classification of UXO with greater confidence.