Full metadata record
| DC Field | Value | Language |
|---|---|---|
| dc.contributor | Department of Land Surveying and Geo-Informatics | en_US |
| dc.contributor.advisor | Shi, Wenzhong (LSGI) | en_US |
| dc.creator | Chen, Shanxiong | - |
| dc.identifier.uri | https://theses.lib.polyu.edu.hk/handle/200/12730 | - |
| dc.language | English | en_US |
| dc.publisher | Hong Kong Polytechnic University | en_US |
| dc.rights | All rights reserved | en_US |
| dc.title | Hierarchical feature analysis for automatic building extraction and change detection | en_US |
| dcterms.abstract | Buildings are the main body and essential elements of a city. The automatic, accurate, and timely acquisition of their location, area, and change attributes is the starting point in many real-world applications. The rapid urbanization process urgently requires timely updating of building information. Building information extraction is significant to smart city construction and sustainable urban development. Building extraction and change detection utilizing high-resolution remote sensing data is the primary approach for updating building information. However, given the diverse building appearances, complex backgrounds, and spectral variability in high-resolution remote sensing data, research on automatic and reliable building extraction and change detection remains challenging in the remote sensing and computer vision community. | en_US |
| dcterms.abstract | The vigorous development of deep learning technology provides a technical basis for automatically and accurately extracting information about buildings and their changes only from high-resolution remote sensing images. However, fully convolutional network-based encoder-decoder models still fail to fully exploit the implicit characteristics of building shapes and the abundant spatial information of the encoder layer, resulting in the loss of spatial information and oversmooth building boundaries. Therefore, a contour guided and local structure-aware encoder-decoder network (CGSANet) is proposed to extract buildings with more accurate boundaries. | en_US |
| dcterms.abstract | Buildings are artificial above-ground objects with regular shapes. The elevation information in light detection and ranging (LiDAR) point clouds can eliminate the spectral confusion of roads and shadow occlusion in images. The abundant spectral and geometric properties of high-resolution images can remove the interference of vegetation in LiDAR data and provide accurate building boundaries. Therefore, this thesis further investigates an automatic building extraction method via multilevel iterative segmentation with LiDAR data and high-resolution imagery fusion, which needs no samples and can improve the degree of automation and robustness of building extraction. | en_US |
| dcterms.abstract | With improved image spatial and temporal resolution, the acquisition of multitemporal high-resolution images provides convenience for building monitoring. Building change detection based on remote sensing data acquired at different times in the same area is an effective and economical approach for updating building information in time. Therefore, this thesis further explores the building change detection method based on the Siamese convolutional network to improve the automation of updating the basic building database. | en_US |
| dcterms.abstract | In summary, in response to the need for rapid updating of building information, this thesis proposes a framework for automatic building extraction and change detection. By fully utilizing the superior feature extraction ability of convolutional neural networks and the complementary characteristics of multisource data, a hierarchical feature analysis method is designed to improve building information extraction quality and automation. The main research contents and innovations are as follows. | en_US |
| dcterms.abstract | (1) A CGSANet is proposed to automatically extract buildings with more accurate boundaries from high-resolution optical remote sensing images. CGSANet is a multitask network composed of a contour-guided (CG) and a multiregion-guided (MRG) module. The CG module is supervised by a building contour that effectively learns building contour-related spatial features to retain the shape pattern of buildings. The MRG module is deeply supervised by four building regions that further capture multiscale and contextual features of buildings. In addition, a hybrid loss function was introduced to guide the model to learn parameters from the pixel-level similarity, local structural similarity, and global similarity to improve the structure learning ability of CGSANet. These three improvements benefit each other synergistically to produce high-quality building extraction results. Experimental results on three challenging public building extraction datasets demonstrate that CGSANet can produce the highest quality results even when extracting buildings with complicated shapes and buildings in images with complex backgrounds. | en_US |
| dcterms.abstract | (2) An unsupervised building extraction method is proposed via multilevel iterative segmentation based on LiDAR data and high-resolution imagery fusion. The method applies multilevel iterative segmentation and hierarchical overlay analysis based on data fusion to automatically extract buildings. A multilevel iterative segmentation method is designed to overcome oversegmentation and undersegmentation based on the global probability of boundary contour detection algorithm. A data fusion-based hierarchical overlay analysis is designed to overcome shadow occlusion. The proposed method achieves competitive building extraction performance without training samples. | en_US |
| dcterms.abstract | (3) A CDasXORNet is proposed for building change detection from bi-temporal remote sensing images by approximating the XOR function. CDasXORNet associates change detection with logical XOR function. A change analysis module is designed by approximating the XOR function based on the bi-temporal features extracted by the Siamese convolutional encoder. The hierarchical XOR operation is performed on hierarchical features to identify building changes. A residual linear attention decoder is introduced by combining residual learning and a linear attention mechanism. Experiments on two building change detection datasets demonstrate that CDasXORNet can effectively mitigate the false detections caused by common challenges in building change detection, such as spectral confusion and building positional inconsistencies between bi-temporal images. CDasXORNet can produce high-quality building change detection results with fewer false alarms and higher overall performance. | en_US |
| dcterms.extent | xx, 214 pages : color illustrations | en_US |
| dcterms.isPartOf | PolyU Electronic Theses | en_US |
| dcterms.issued | 2023 | en_US |
| dcterms.educationalLevel | Ph.D. | en_US |
| dcterms.educationalLevel | All Doctorate | en_US |
| dcterms.LCSH | Buildings -- Location -- Remote sensing | en_US |
| dcterms.LCSH | Machine learning | en_US |
| dcterms.LCSH | Optical radar | en_US |
| dcterms.LCSH | Image processing -- Digital techniques | en_US |
| dcterms.LCSH | Hong Kong Polytechnic University -- Dissertations | en_US |
| dcterms.accessRights | open access | en_US |
Copyright Undertaking
As a bona fide Library user, I declare that:
- I will abide by the rules and legal ordinances governing copyright regarding the use of the Database.
- I will use the Database for the purpose of my research or private study only and not for circulation or further reproduction or any other purpose.
- I agree to indemnify and hold the University harmless from and against any loss, damage, cost, liability or expenses arising from copyright infringement or unauthorized usage.
By downloading any item(s) listed above, you acknowledge that you have read and understood the copyright undertaking as stated above, and agree to be bound by all of its terms.
Please use this identifier to cite or link to this item:
https://theses.lib.polyu.edu.hk/handle/200/12730