Previous MSc Theses

"A Patch-Based Deep Learning Approach for Land-Classification of Sentinel-2 Satellite Imagery." D. Smith. M. Roper. Department of Computer and Information Sciences, University of Strathclyde. 2019. Download PDF (BibTeX) ACSBD

Abstract:
The application of machine learning to the remote sensing field has produced many exciting new advancements. One of these is in land-classification: identifying the land-cover content present in satellite imagery. The use of deep neural networks has provided stateof-the-art accuracies for land-cover classification. This project experiments with deep learning architectures and novel image handling to methods to improve the classification accuracies achievable by Global Surface Intelligence, a geospatial analytics company. By implementing segmentation and patch-based training, we successfully show that by expanding their current methodology to consider spatial content, they can achieve increased classification accuracies.