Date of Award

2023

Document Type

Thesis

Degree Name

Master of Science (MS)

Department

Computer Science

Committee Chair

Tathagata Mukherjee

Committee Member

Aaron Kaulfus

Committee Member

Chaity Banerjee Mukherjee

Subject(s)

Big data, Machine learning, Remote-sensing images--Classification

Abstract

Advanced Deep Learning techniques have been widely used in Spatial Data Science specifically with Geographic Information Systems (GIS). Petabyte scale data production is occurring due to advances in remote sensing capabilities within the commercial and government sectors. Discovering relevant data for science research and applications can become difficult with large volumes of spatially and temporally diverse data availability. The NASA Commercial Smallsat Data Acquisition (CSDA) Program acquires and archives data from commercial satellite remote sensing vendors. While the underlying observation characteristics may be similar, such as being visible imagery, varying orbital and acquisition strategies result in a disparate data archive. This paper applies Simsiam (Simple Siamese Representation Learning) model along with the Transformer as self-supervised machine learning techniques to address the following challenges in discovering satellite imagery available through the CSDA Program: 1) the archive has a large amount of data and labeling each of scene manually for the features they contain would be tedious and time consuming, 2) Each of the vendor satellites have different instruments, orbital characteristics, and therefore different spatial, temporal, and radiometric resolutions, 3) For a given event a particular vendor may have data while others may not, and, 4) end users desire a single API or interface for the data extraction from different satellite vendors.

Available for download on Wednesday, December 11, 2024

Share

COinS