| Campo DC | Valor | Idioma |
|---|
| dc.contributor.advisor | Carvalho Júnior, Osmar Abílio de | - |
| dc.contributor.author | Castro, Heitor da Rocha Nunes de | - |
| dc.date.accessioned | 2024-06-12T17:34:17Z | - |
| dc.date.available | 2024-06-12T17:34:17Z | - |
| dc.date.issued | 2024-06-12 | - |
| dc.date.submitted | 2023-06-16 | - |
| dc.identifier.citation | CASTRO, Heitor da Rocha Nunes de. Detecção de depressões cársticas no Brasil usando segmentação semântica e de instância e comparando diferentes modelos digitais globais de elevação. 2023. 59 f., il. Dissertação (Mestrado em Geografia) - Universidade de Brasília, Brasília, 2023. | pt_BR |
| dc.identifier.uri | http://repositorio2.unb.br/jspui/handle/10482/48233 | - |
| dc.description | Dissertação (mestrado) - Universidade de Brasília, Instituto de Ciências Humanas, Departamento de Geografia, Programa de Pós-Graduação em Geografia, 2023. | pt_BR |
| dc.language.iso | eng | pt_BR |
| dc.rights | Acesso Aberto | pt_BR |
| dc.title | Detecção de depressões cársticas no Brasil usando segmentação semântica e de instância e comparando diferentes modelos digitais globais de elevação | pt_BR |
| dc.type | Dissertação | pt_BR |
| dc.subject.keyword | Inteligência artificial | pt_BR |
| dc.subject.keyword | Geomorfologia | pt_BR |
| dc.subject.keyword | Dolina | pt_BR |
| dc.subject.keyword | Deep learning | pt_BR |
| dc.rights.license | A concessão da licença deste item refere-se ao termo de autorização impresso assinado pelo autor com as seguintes condições: Na qualidade de titular dos direitos de autor da publicação, autorizo a Universidade de Brasília e o IBICT a disponibilizar por meio dos sites www.unb.br, www.ibict.br, www.ndltd.org sem ressarcimento dos direitos autorais, de acordo com a Lei nº 9610/98, o texto integral da obra supracitada, conforme permissões assinaladas, para fins de leitura, impressão e/ou download, a título de divulgação da produção científica brasileira, a partir desta data. | pt_BR |
| dc.description.abstract1 | This research aims to investigate the use of deep segmentation in detecting and quantifying natural karst depressions developed in the carbonate rocks of the Neoproterozoic Bambuí Group in Western Bahia, Brazil. The karst landscape of the study
area has dolines and the formation of lakes enclosed in limestone. The study analyzes different approaches to detecting karst depressions. First, a comparison of five different Global Digital Terrain Models (DEM) with 30 meters resolution: Copernicus 30m Global DEM (GLO-30), ALOS World 3D (AW3D30), Shuttle Radar Topography Mission (SRTM), National Aeronautics and Space Administration DEM (NASADEM), and Advanced Spaceborne Thermal Emission and Reflection Radiometer - Global DEM
(ASTER-GDEM). Second, comparing five semantic segmentation architectures with EfficientNet-B7 backbone (Feature Pyramid Network - FPN, LinkNet, Unet, Unet++, and DVL3+) and one instance segmentation (Mask-RCNN). Third, evaluation of
segmentation elaboration using two variables (DEM and DEM-based sink depth) or eleven variables (DEM, DEM-based sink depth, and nine terrain attributes). The research did not evaluate the use of DEM in isolation due to its very low accuracy in previous analyses. The methodology had the following steps: (a) acquisition of DEMs and generation of geomorphometric attributes; (b) sample labeling by manual interpretation of karst depressions from Sentinel-2 and OLI-Landsat 8 images; (c) selection of samples for training (1600 samples), validation (400 samples) and testing (400 samples) with dimensions 128x128 considering two channels (DEM and depth of sinking based on DEM) and eleven channels (the two previous ones added by nine morphometric attributes); (e) elaboration of semantic and instance segmentations; (f) accuracy analysis; (g) image reconstruction using sliding window; and (f) conversion from semantic segmentation to instance using GIS tools. The results show that the GLO-30 data showedthe highest accuracy values, followed by the AW3D30. In contrast, the ASTER GDEM obtained the worst results. Among the models using semantic segmentation, the FPN presented the most significant accuracy results, while the DVL3+ presented the worst. Considering the same architectures and DEM, the models that used 11 channels obtained better results than those that used only two channels. Converting data from semantic segmentation to instance segmentation using a GIS tool proved to be very easy since the features did not interact. | pt_BR |
| dc.description.unidade | Instituto de Ciências Humanas (ICH) | pt_BR |
| dc.description.unidade | Departamento de Geografia (ICH GEA) | pt_BR |
| dc.description.ppg | Programa de Pós-Graduação em Geografia | pt_BR |
| Aparece nas coleções: | Teses, dissertações e produtos pós-doutorado
|
