Convolutional Neural Networks for Road Detection: An Unsupervised Domain Adaptation Approach

Registro completo de metadados
MetadadosDescriçãoIdioma
Autor(es): dc.contributorUniversidade Estadual Paulista (UNESP)-
Autor(es): dc.contributorBrazilian Army Geographic Service-
Autor(es): dc.contributorPurdue University-
Autor(es): dc.creatorCollegio, Gustavo Rota-
Autor(es): dc.creatorDal Poz, Aluir Porfírio-
Autor(es): dc.creatorFilho, Antonio Gaudencio Guimarães-
Autor(es): dc.creatorHabib, Ayman-
Data de aceite: dc.date.accessioned2025-08-21T21:55:47Z-
Data de disponibilização: dc.date.available2025-08-21T21:55:47Z-
Data de envio: dc.date.issued2025-04-29-
Data de envio: dc.date.issued2024-06-11-
Fonte completa do material: dc.identifierhttp://dx.doi.org/10.5194/isprs-archives-XLVIII-2-2024-65-2024-
Fonte completa do material: dc.identifierhttps://hdl.handle.net/11449/309852-
Fonte: dc.identifier.urihttp://educapes.capes.gov.br/handle/11449/309852-
Descrição: dc.descriptionDue to the frequent road network changes, keeping them updated is fundamental for several purposes. Currently, models based on Deep Learning (DL), specifically, Convolutional Neural Networks (CNNs), such as encoder-decoder type, are state-of-the-art for this purpose. In this context, the high performance in CNNs has two aspects involved: the model needs a large labeled dataset, and the dataset belongs to the same probability distribution. In practical applications, however, this may not hold, since there is a domain shift effect, and it is not customary for the availability of labeled data. To approach these challenges, we propose to adapt the U-Net architecture (encoder-decoder) to the Unsupervised Domain Adaptation (UDA) that does not need labeling data to minimize the domain shift effect. Our results demonstrate that the proposed method contributes to road segmentation, whose model reaches 74.31% (IoU) and 85.04% (F1), against the same model without UDA that reaches 67.36% (IoU) and 80.02% (F1). This implies that the information that comes from the target domain, even unsupervised, contributes to adversarial learning, improving the generalization capacity of the model, enhancing aspects such as better discrimination surrounding classes, and in the geometric delineation of the road network.-
Descrição: dc.descriptionCoordenação de Aperfeiçoamento de Pessoal de Nível Superior (CAPES)-
Descrição: dc.descriptionFundação de Amparo à Pesquisa do Estado de São Paulo (FAPESP)-
Descrição: dc.descriptionDepartment of Cartography Faculty of Sciences and Technology São Paulo State University (UNESP)-
Descrição: dc.descriptionDSG Brazilian Army Geographic Service, DF-
Descrição: dc.descriptionLyles School of Civil Engineering Purdue University-
Descrição: dc.descriptionDepartment of Cartography Faculty of Sciences and Technology São Paulo State University (UNESP)-
Descrição: dc.descriptionFAPESP: 2021/03586-2-
Formato: dc.format65-71-
Idioma: dc.languageen-
Relação: dc.relationInternational Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences - ISPRS Archives-
???dc.source???: dc.sourceScopus-
Palavras-chave: dc.subjectAdversarial Training-
Palavras-chave: dc.subjectDomain Adaptation-
Palavras-chave: dc.subjectHigh-Resolution Images-
Palavras-chave: dc.subjectRoad Detection-
Palavras-chave: dc.subjectSemantic Segmentation-
Título: dc.titleConvolutional Neural Networks for Road Detection: An Unsupervised Domain Adaptation Approach-
Tipo de arquivo: dc.typeaula digital-
Aparece nas coleções:Repositório Institucional - Unesp

Não existem arquivos associados a este item.