Date on Master's Thesis/Doctoral Dissertation

5-2023

Document Type

Master's Thesis

Degree Name

M.S.

Department

Computer Engineering and Computer Science

Degree Program

Computer Science, MS

Committee Chair

Frigui, Hichem

Committee Member

Karem, Andrew

Committee Member

Baidya, Sabur

Author's Keywords

Deep learning; data augmentation; automatic target recognition; neural networks; yolo; CNN

Abstract

Automatic Target Recognition (ATR) is a task that aims to recognize targets based on data obtained from visual sensors such as Infrared cameras. Machine learning, in particular Deep Learning, has been the dominate approach to solve this task. However, in order to achieve high performance, Deep Learning requires a large amount of data that may not be available, which poses an issue when developing robust models. In this thesis, we propose the use of data augmentation from an automatic detector as an alternative source to supplement the need for more training data. We use a two stage approach to develop an ATR. In the first stage, we use a Yolo detector to capture regions of interest. In the second stage, we use a classifier to predict target types. We use yolo detections as augmentations where we consider detections with high confidence as labeled and detections with low confidence as unlabeled. When only labeled data is used we learn a fully-supervised Convolutional Neural Network classifier and compare the performance of the model with and without augmentation. When both labeled and unlabeled data are used, we rely on a Semi-Supervised classifier and compare its performance as we vary the labeled and unlabeled data. Through our experiments we show that by using yolo detections for data augmentation we are able to improve the accuracy of the ATR.

Download

Share

COinS