A lightweight building instance extraction method based on adaptive optimization of mask contour
Automatic extraction of building instances from high spatial resolution optical remote sensing imagery is essential for urban infrastructure and smart management. In view of the severe challenges such as intricate building samples, high training overhead and inaccurate mask contours, this paper prop...
| Main Authors: | , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Elsevier
2023-08-01
|
| Series: | International Journal of Applied Earth Observations and Geoinformation |
| Subjects: | |
| Online Access: | http://www.sciencedirect.com/science/article/pii/S1569843223002443 |
| _version_ | 1797738003785842688 |
|---|---|
| author | Xiaoxue Liu Yiping Chen Cheng Wang Kun Tan Jonathan Li |
| author_facet | Xiaoxue Liu Yiping Chen Cheng Wang Kun Tan Jonathan Li |
| author_sort | Xiaoxue Liu |
| collection | DOAJ |
| description | Automatic extraction of building instances from high spatial resolution optical remote sensing imagery is essential for urban infrastructure and smart management. In view of the severe challenges such as intricate building samples, high training overhead and inaccurate mask contours, this paper proposes a region-based, two-stage segmentation model assembled with components of adaptive feature extraction, guided anchoring and iterative subdivision mask, resulting in lightweight extraction of building instances and adaptive optimization of mask contours. We conducted comparison experiments with several classical, state-of-the-art instance segmentation methods on WHU aerial dataset, China city satellite dataset and Inria aerial dataset. The quantitative evaluation demonstrates that our method signally reduces the computational load by at least 34.5%, increases the mask AP on the three datasets by at least 0.8%, 4.6% and 4.1%, respectively, and increases the mean performance (mPC) and relative performance (rPC) under image corruption (12 corruptions and 2 severity levels) by at least 2.4% and 3.2%, respectively. The qualitative evaluation further verifies that our method is more discriminative for features such as color, texture, and shadow, and more adaptable to changes in shape, scale, and distribution and so on. Especially for large-sized buildings, various comparison methods either miss the partial interior area or excessively smooth the mask contour, whereas our method can output complete and accurate masks. |
| first_indexed | 2024-03-12T13:37:37Z |
| format | Article |
| id | doaj.art-b583fe504c944bb880e6d862491221fe |
| institution | Directory Open Access Journal |
| issn | 1569-8432 |
| language | English |
| last_indexed | 2024-03-12T13:37:37Z |
| publishDate | 2023-08-01 |
| publisher | Elsevier |
| record_format | Article |
| series | International Journal of Applied Earth Observations and Geoinformation |
| spelling | doaj.art-b583fe504c944bb880e6d862491221fe2023-08-24T04:34:15ZengElsevierInternational Journal of Applied Earth Observations and Geoinformation1569-84322023-08-01122103420A lightweight building instance extraction method based on adaptive optimization of mask contourXiaoxue Liu0Yiping Chen1Cheng Wang2Kun Tan3Jonathan Li4Fujian Key Laboratory of Sensing and Computing for Smart Cities, School of Informatics, Xiamen University, Xiamen, China; School of Big data and Computer Science, Guizhou Normal University, Guiyang, ChinaSchool of Geospatial Engineering and Science, Sun Yat-sen University, Zhuhai, China; Corresponding authors.Fujian Key Laboratory of Sensing and Computing for Smart Cities, School of Informatics, Xiamen University, Xiamen, ChinaEast China Normal University, Key Laboratory of Spatial-Temporal Big Data Analysis and Application of Natural Resources in Megacities (Ministry of Natural Resources), Shanghai, ChinaDepartment of Geography and Environmental Management and Department of Systems Design Engineering, University of Waterloo, Waterloo, Canada; Corresponding authors.Automatic extraction of building instances from high spatial resolution optical remote sensing imagery is essential for urban infrastructure and smart management. In view of the severe challenges such as intricate building samples, high training overhead and inaccurate mask contours, this paper proposes a region-based, two-stage segmentation model assembled with components of adaptive feature extraction, guided anchoring and iterative subdivision mask, resulting in lightweight extraction of building instances and adaptive optimization of mask contours. We conducted comparison experiments with several classical, state-of-the-art instance segmentation methods on WHU aerial dataset, China city satellite dataset and Inria aerial dataset. The quantitative evaluation demonstrates that our method signally reduces the computational load by at least 34.5%, increases the mask AP on the three datasets by at least 0.8%, 4.6% and 4.1%, respectively, and increases the mean performance (mPC) and relative performance (rPC) under image corruption (12 corruptions and 2 severity levels) by at least 2.4% and 3.2%, respectively. The qualitative evaluation further verifies that our method is more discriminative for features such as color, texture, and shadow, and more adaptable to changes in shape, scale, and distribution and so on. Especially for large-sized buildings, various comparison methods either miss the partial interior area or excessively smooth the mask contour, whereas our method can output complete and accurate masks.http://www.sciencedirect.com/science/article/pii/S1569843223002443Building instance extractionRemote sensing imageryDeep learningMask contourAdaptive and lightweight |
| spellingShingle | Xiaoxue Liu Yiping Chen Cheng Wang Kun Tan Jonathan Li A lightweight building instance extraction method based on adaptive optimization of mask contour International Journal of Applied Earth Observations and Geoinformation Building instance extraction Remote sensing imagery Deep learning Mask contour Adaptive and lightweight |
| title | A lightweight building instance extraction method based on adaptive optimization of mask contour |
| title_full | A lightweight building instance extraction method based on adaptive optimization of mask contour |
| title_fullStr | A lightweight building instance extraction method based on adaptive optimization of mask contour |
| title_full_unstemmed | A lightweight building instance extraction method based on adaptive optimization of mask contour |
| title_short | A lightweight building instance extraction method based on adaptive optimization of mask contour |
| title_sort | lightweight building instance extraction method based on adaptive optimization of mask contour |
| topic | Building instance extraction Remote sensing imagery Deep learning Mask contour Adaptive and lightweight |
| url | http://www.sciencedirect.com/science/article/pii/S1569843223002443 |
| work_keys_str_mv | AT xiaoxueliu alightweightbuildinginstanceextractionmethodbasedonadaptiveoptimizationofmaskcontour AT yipingchen alightweightbuildinginstanceextractionmethodbasedonadaptiveoptimizationofmaskcontour AT chengwang alightweightbuildinginstanceextractionmethodbasedonadaptiveoptimizationofmaskcontour AT kuntan alightweightbuildinginstanceextractionmethodbasedonadaptiveoptimizationofmaskcontour AT jonathanli alightweightbuildinginstanceextractionmethodbasedonadaptiveoptimizationofmaskcontour AT xiaoxueliu lightweightbuildinginstanceextractionmethodbasedonadaptiveoptimizationofmaskcontour AT yipingchen lightweightbuildinginstanceextractionmethodbasedonadaptiveoptimizationofmaskcontour AT chengwang lightweightbuildinginstanceextractionmethodbasedonadaptiveoptimizationofmaskcontour AT kuntan lightweightbuildinginstanceextractionmethodbasedonadaptiveoptimizationofmaskcontour AT jonathanli lightweightbuildinginstanceextractionmethodbasedonadaptiveoptimizationofmaskcontour |