Georgia Southern University
Application of Semantic Segmentation to an Automatic Target Recognition Problem
Abstract
dc:description.abstractSynthetic Aperture Radar (SAR) is an active remote sensing system commonly used in aerial reconnaissance. SAR penetrates cloud cover and vegetation by recording reflected energy pulses. The resulting information content is difficult to interpret due to vast clutter data and sparse target data. The time-consuming process of analyzing SAR imagery can be greatly reduced by implementing an Automatic Target Recognition (ATR) algorithm. Convolutional Neural Networks (CNN) are capable of extracting identification features from SAR data content. Further research in this field indicates that high performing models are insufficiently robust due to high clutter correlation among classes in the Moving and Stationary Target Acquisition and Recognition (MSTAR) Mixed Target dataset. This research proposes a robust classification pipeline with the addition of a semantic segmentation model. The result is a classification model that is trained to extract identification features from only target and shadow pixels in the MSTAR Mixed Target dataset.
Degree
thesis:*- Name thesis:degree_name
- Master of Science in Mathematics (M.S.)
- Level thesis:degree_level
- Thesis (open access)
- Discipline thesis:degree_discipline
- Department of Mathematical Sciences
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Rush, Bruce W., Jr.
- Contributors dc:contributor
-
- Scott Kersey
- Yan Wu
Subjects
dc:subject × 4Rights
- Licence dc:rights.license
Identifiers
dc:identifier.*- Repository record dc:identifier
- https://digitalcommons.georgiasouthern.edu/etd/2954