Classification Study using a Handheld, Three-Element EMI Sensor
Abstract
Typical library matching procedures used for classification ask how much a given target's principal axis polarizabilities resemble those of targets of interest (TOIs). The polarizabilities are calculated from sensor array data using dipole inversion. In conventional dipole inversion, given a set of measurements of the EMI response we simultaneously search out the location orientation, and principal axis polarizabilities that produce signals that best match the measured target response. Unfortunately you need a lot of data to reliably constrain the target location in dipole inversion. The eigenvalues of the data matrix for an array of paired transmit and receive loop are easy to compute and need a much less extensive sensor array than the polarizabilities. This research project has focused on whether or not the data eigenvalues from a simple handheld sensor can serve as surrogates for principal axis polarizabilities. We conclude that they cannot. In general, the data eigenvalues are built up of linear combinations of the principal axis polarizabilities. This means that a classification scheme based on data eigenvalues would have to accommodate arbitrary linear combinations of TOI polarizabilities. The loss of classification performance to clutter items whose principal axis polarizabilities can be closely reproduced using linear combinations of TOI polarizabilities is just too large.
Document Details
- Document Type
- Technical Report
- Publication Date
- Apr 01, 2013
- Accession Number
- ADA600400
Entities
People
- Thomas Bell
Organizations
- Leidos