University of Twente Student Theses
Deep Learning Framework for Urban Impervious Surface Mapping Using Open Multisource Geospatial Data
Barathidhasan, Shanmathi (2023) Deep Learning Framework for Urban Impervious Surface Mapping Using Open Multisource Geospatial Data.
PDF
2MB |
Abstract: | Developing a comprehensive understanding of urbanization and the urban ecosystem necessitates the requirement of crucial knowledge pertaining to Urban Impervious Surfaces (UIS). The need for an updated and accurate classification of UIS is of growing importance. Numerous studies have utilized various remote sensing and geospatial data sources to address this need. In this study, we investigate the effective utilization of freely available remote sensing sources, including satellite imagery and Volunteered Geographic Information sources like Twitter and Open Street Map, for UIS classification. An innovative aspect of this research lies in the creative integration of these data sources to acquire UIS training labels required for a Deep Learning based classification system. We propose a one-class classification approach using Deep One-Class Classification (DOCC) to reduce dependency on labeled data for different classes for an effective UIS classification. The introduced DOCC model demonstrates the efficiency of a deep feature network, utilizing only limited spectral features such as blue, green, and red spectral bands, to achieve accurate UIS classification. The DOCC showed a good accuracy which was later compared with that of a deep multiclass classification, focusing only on UIS class of multiclass. Moreover, the UIS labels proved to be more efficient in comparison to that of the globally available impervious maps. Overall, this study aims in contributing to the advancement of UIS mapping techniques by integrating various geospatial data sources and exploring innovative classification techniques. Keywords: Sentinel 2, Volunteered Geographic Information, Twitter, Open Street Map, Urban Impervious Surface, One-Class Classification, Deep feature learning. |
Item Type: | Essay (Master) |
Faculty: | ITC: Faculty of Geo-information Science and Earth Observation |
Subject: | 38 earth sciences, 43 environmental science, 54 computer science |
Programme: | Geoinformation Science and Earth Observation MSc (75014) |
Link to this item: | https://purl.utwente.nl/essays/95777 |
Export this item as: | BibTeX EndNote HTML Citation Reference Manager |
Repository Staff Only: item control page