Title

Radar Returns with Artificial Neural Networks

Document Type

Thesis

Degree Name

Master of Science (MS)

Department

Computer Science and Info Sys

Date of Award

Fall 2020

Abstract

The research discussed in this project is based on previous results that examined the use of neural networks to classify radar returns (Sigillito et al., 1989). In particular, in the field of radar modeling used by the Missile Defense Agency (MDA), one must accurately model aspects of the radar systems, including the complex algorithms for the signal processing of radar returns, to accurately classify radar return signals. Previous results demonstrated that radar returns from the ionosphere could be adequately classified using machine learning techniques such as neural networks. In this research, we looked at improving the classification results of such radar modeling. We compared the performance of a single layer with multi-layer feed-forward networks, and looked at the effects of modifying various meta-parameters of such networks, such as varying hidden nodes and learning regularization parameters on the performance of the machine classifiers. In addition, data augmentation techniques were explored in an attempt to expand the data set and improve the trained classifiers. We also applied more recent deep-learning network techniques to the classification of such radar returns, to see if more modern architectures and methods can significantly improve the model’s performance. We demonstrated that in fact recurrent network architectures can improve on past best performance reported for this task.

Advisor

Derek Harter

Subject Categories

Computer Sciences | Physical Sciences and Mathematics

Link to Full Text
COinS