Show simple item record

dc.contributor.authorWu, Yifan 14:27:26 (GMT) 14:27:26 (GMT)
dc.description.abstractBuilding extraction from remote sensing images is a critical task to support various applications such as cartography, disaster response, and urban planning. The automation of this task is an active research area due to the time-consuming nature and high expense associated with the manual approach. However, traditional computer vision methods rely on handcrafted features and human knowledge, leading to the lack of the ability to leverage big remote sensing data. Although recently developed deep learning based methods brought significant advancements in the identification and coarse annotation of buildings, the accuracy and precision of extracted buildings are still insufficient for high-precision applications such as surveying and mapping. This thesis presents two works aiming at enhanced building extraction from high-resolution remote sensing images by tackling key issues in building footprint extraction and building vectorization. For building footprint extraction, to address the heterogeneous noisy features around building boundaries, this thesis presents a deep learning strategy that incorporates a topography-aware loss (TAL) within a multi-resolution fusion architecture to increase the accuracy of boundaries in building segmentation. For building vectorization, to address the interference caused by noise and obstruction from shadows and trees around buildings and the limited receptive field in deep learning networks, this thesis presents a framework that combines a deep learning based building edge detection strategy and a geometry-guided building polygon reconstruction method for improved building outline vectorization in terms of vertex accuracy. Comparative experimental results on high-resolution remote sensing building datasets demonstrate significant improvements in building boundary accuracy and polygon vertex accuracy respectively over state-of-the-art methods. Hence, both works provide new means to address challenges posed by complex environmental conditions around buildings captured in remote sensing images and enable accurate building segmentation and vectorization towards automatic building extraction for high-precision applications.en
dc.publisherUniversity of Waterlooen
dc.subjectdeep learningen
dc.subjectbuilding extractionen
dc.subjectremote sensing imagesen
dc.titleDeep Learning Based Building Extraction from High-Resolution Remote Sensing Imagesen
dc.typeMaster Thesisen
dc.pendingfalse Design Engineeringen Design Engineeringen of Waterlooen
uws-etd.degreeMaster of Applied Scienceen
uws.contributor.advisorClausi, David
uws.contributor.advisorXu, Linlin
uws.contributor.affiliation1Faculty of Engineeringen

Files in this item


This item appears in the following Collection(s)

Show simple item record


University of Waterloo Library
200 University Avenue West
Waterloo, Ontario, Canada N2L 3G1
519 888 4883

All items in UWSpace are protected by copyright, with all rights reserved.

DSpace software

Service outages