Date of Award

12-1994

Document Type

Thesis

Degree Name

Master of Science in Electrical Engineering

Department

Department of Electrical and Computer Engineering

First Advisor

Steven K. Rogers, PhD

Abstract

This thesis examines the discrimination of targets with Ultra High Range Resolution (UHRR) radar data. Using these measured signals from frontal aspect angles of four aircraft classes, the baseline performance of the Adaptive Gaussian Classifier (AGC) is tested with respect to aligning exemplars to templates. Alignment plays a crucial role in the AGC's classification performance which can degrade by 11% for a target class. The AGC is compared to non-parametric classifiers, but no statistically significant degradation of performance is found. Data separability is analyzed by hounding the Bayes error. The data is well separated in a statistical sense. A feature selection algorithm based on analysis of the decision boundary, is applied to find a reduced feature set, which are linear combinations of the original features. These features are optimized with respect to classification error rather than reconstruction error. This technique is extended to deduce the relevant features in the original feature space. Fewer than 5% of the features in the original feature space may be used to attain an improved classification rate. This new method is a true reduction of features and shows improvement up to 15%. Discrimination of UHRR radar signatures using a multiresolution analysis is proposed. The decision boundary analysis chooses relevant wavelet scales with respect to classification. Some improved performance against an entropy based measure is observed for limited feature sets. The technique developed here successfully chooses the scale that causes classification performance to peak within 5% of the performance in the full-dimensional or reduced-dimensional UHRR radar signature space.

AFIT Designator

AFIT-GE-ENG-94D-07

DTIC Accession Number

ADA289378

Comments

The author's Vita page is omitted.

Download

Share

COinS