Date of Award


Document Type


Degree Name

Master of Science


Department of Electrical and Computer Engineering

First Advisor

Steven K. Rogers, PhD


The Feature Space Trajectory Neural Network (FSTNN) is a simple yet powerful pattern recognition tool developed by Neiberg and Casasent for use in an Automatic Target Recognition System. Since the FSTNN was developed, it has been used on various problems including speaker identification and space object identification. However, in these types of problems, the test set represents time series data rather than an independent set of points. Since the distance metric of the standard FSTNN treats each test point independently without regard to its position in the sequence, the FSTNN can yield less than optimal results in these problems. Two methods for incorporating sequence information into the FSTNN algorithm are presented. These methods, Dynamic Time Warping (DTW) and Uniform Time Warping (UTW), are described and compared to the standard FSTNN performance on the space object identification problem. Both reduce error induced by improper synchronization of the test and training sequences and make the FSTNN more generally applicable to a wide variety of pattern recognition problems. They incorporate sequencing information by synchronizing the test and training trajectories. DTW accomplishes this 'on-the-fly' as the sequence progresses while UTW uniformly compensates for temporal differences across the trajectories. These algorithms improve the maximum probability of false alarm (PFA) of the standard FSTNN by an average of 10.18% and 27.69%, respectively, although UTW is less consistent in its results. A metric for determining the saliency of the features in an FSTNN is also presented and demonstrated.

AFIT Designator


DTIC Accession Number