Tensor Invariant Processing of Multistatic EMI Data for Target Classification
Dr. Thomas Bell | Leidos
The objective of this project is to test and evaluate a new procedure for target classification in the context of the ESTCP series of live‐site classification demonstrations. The procedure is transparent, objective, and easily automated. Classification decisions are based on two parameters easily calculated from magnetic polarizabilities of unknown targets and targets of interest (TOI). The parameters are measures of the mismatch between the strength and the shape of the respective polarizability curves. Test results using data from the recent Camp Beale and Pole Mountain demonstrations show that this procedure can produce improved classification performance over conventional processing.
The processing and analysis procedures to be demonstrated were developed in SERDP projects MR‐1658 and MR‐2100 for use with the advanced electromagnetic induction (EMI) arrays developed by SERDP and ESTCP for target classification. The classification algorithm exploits the fact that an object’s polarizability is a product of two factors: the volume of the object and a tensor whose eigenvalues depend only on the shape and composition of the object. Confronted with an unknown target, this project seeks to compare its apparent size and EMI “shape” with the sizes and shapes of TOI. Classification is based on thresholding a figure of merit (FOM) parameter that is a weighted sum of parameters quantifying the mismatches in the EMI size and shape of the target relative to the TOI. For multiple TOI, the FOM is minimized over the set of TOI. This basic algorithm is also used in cluster analysis to identify unexpected munitions. In this case each target is compared against all others to find groups which have similar EMI size and shape. A novel procedure for extracting polarizabilities from multi‐axis, multi‐static data collected with the new generation of transient EMI (TEM) sensors will also be tested. The procedure is based on sampling various sets of three transmit and three receive elements to calculate independent estimates of rotational invariants of the polarizability tensor. It has been shown to produce more accurate polarizabilities than conventional dipole inversion in some cases, and provides a direct measure of the uncertainty in polarizability estimates.
Benefits of this technology are both quantitative and qualitative. Re‐processing data from the recent Camp Beale demonstration using this approach produced a Receiver Operating Characteristic (ROC) curve that rises more rapidly and hits the 100% TOI recovered level with 50% fewer clutter digs beyond the training set than the ROC curve from conventional processing. Improved classification performance improves munitions response efficiency. The procedure operates in an intuitive and easily visualized feature space. It is transparent, objective, and easily automated. All of this is likely to facilitate transition to production work and ease regulatory acceptance. (Anticipated Project Completion - 2015)
Points of Contact
Dr. Thomas Bell
SERDP and ESTCP