f Classification of UXO By Principal Dipole Polarizability

Data acquired by multiple-transmitter, multiple-receiver time-domain electromagnetic instruments show great potential for determining the geometry and composition of near surface conductive targets. Here is presented analysis of data from one such system; the Berkeley Unexploded-ordnance Discriminator (BUD) system. BUD multi-static data matrices data are succinctly reduced to obtain magnetic dipole polarizability tensors for each time gate. When viewed over all time gates, the projections of the data onto the principal polar axes yield so-called polarizability curves. these curves are especially well suited to discriminating anomalies corresponding to objects of rotational symmetry from those of irregularly shaped objects. the curves have previously been successfully employed as library elements in a pattern recognition scheme aimed at discriminating harmless scrap metal from dangerous intact unexploded ordnance. However, previous polarizability-curve matching methods have only been applied at field sites which are known a priori to be contaminated by a single type of ordnance, and furthermore, the particular ordnance present in the subsurface was known to be large. Thus signal amplitude was a key element in the discrimination process. the work presented here applies feature-based pattern classification techniques to BUD field data where more than 20 categories of object are present, spanning a broad range of sizes. Data from a calibration grid at the Yuma, AZ proving ground are used in a cross validation study to calibrate the pattern recognition method. the resultant method is then applied to the Yuma Blind Test Grid data. Results indicate that when lone UXO are present and SNR is reasonably high, Polarizability Curve Matching successfully discriminates UXO from scrap metal when a broad range of objects are present.