Date of Award

12-1997

Document Type

Thesis

Degree Name

Master of Science

Department

Department of Electrical and Computer Engineering

First Advisor

Andrew J. Terzuoli, Jr., PhD

Abstract

The dispersive scattering center (DSC) model characterizes high-frequency backscatter from radar targets as a finite sum of localized scattering geometries distributed in range, these geometries, along with their relative locations, can be conveniently used as features in a one-dimensional automatic target recognition (ATR) algorithm. The DSC model's type and range parameters correspond to geometry and distance features according to the geometric theory of diffraction (GTD). Since these parameters are estimated in the phase history domain of the radar signal, the range parameter does provide superresolution in the time domain. To demonstrate the viability of feature extraction based on the DSC model's range and type parameters, a four class ATR experiment was performed. The experimental data contains 301 direct range measurements each for four model aircraft of similar size and shape at 0 degrees elevation and from 0 to 30 degrees azimuth. After implementing DSC model feature extraction on this data, a fully-connected, two-layer neural net obtained over 98% classification accuracy. In addition, DSC model feature extraction offers an approximate 85% reduction in the number of features compared to the numerous Fourier bin magnitudes in template matching approaches to ATR.

AFIT Designator

AFIT-GE-ENG-97D-05

DTIC Accession Number

ADA335655

Download

Share

COinS