Publicação
Texture Analysis of SPOT5 Data for Land Cover Mapping on the Metropolitan Area of Lisbon
| Resumo: | This paper argues that the integration of texture analysis in image classification may help to increase the overall accuracy in urban cartography, by reducing the problem of spectral variability. In a first phase, images were classified with the following algorithms: Parallelepiped, Minimum Distance, Mahalanobis Distance, Maximum Likelihood, Spectral Angle Mapper, Binary Encoding, and Neural Net. The best results in each land cover class were selected, producing two pre-final maps, one with urban classes and another with non urban. Texture analyses were applied in the first result for detecting and correction of errors in the residential class. In a second phase the following image transformations were implemented: Minimum Noise Fraction, Spectral Unmixing, and Mixture Tuned Matched Filtering. This step allowed the production of masks for the different urban classes. An increase in the overall accuracy occurred between the first and second phases, namely form the Maximum Likelihood classifier (North image 73%, South image 76%) to around 80% in the final mapping. |
|---|---|
| Autores principais: | Gaspar, Nuno |
| Outros Autores: | Tenedório, José A.; Santos, Teresa; Rocha, Jorge |
| Assunto: | Remote Sensing Algorithms |
| Ano: | 2010 |
| País: | Portugal |
| Tipo de documento: | capítulo de livro |
| Tipo de acesso: | acesso restrito |
| Instituição associada: | Universidade de Lisboa |
| Idioma: | inglês |
| Origem: | Repositório da Universidade de Lisboa |
| Resumo: | This paper argues that the integration of texture analysis in image classification may help to increase the overall accuracy in urban cartography, by reducing the problem of spectral variability. In a first phase, images were classified with the following algorithms: Parallelepiped, Minimum Distance, Mahalanobis Distance, Maximum Likelihood, Spectral Angle Mapper, Binary Encoding, and Neural Net. The best results in each land cover class were selected, producing two pre-final maps, one with urban classes and another with non urban. Texture analyses were applied in the first result for detecting and correction of errors in the residential class. In a second phase the following image transformations were implemented: Minimum Noise Fraction, Spectral Unmixing, and Mixture Tuned Matched Filtering. This step allowed the production of masks for the different urban classes. An increase in the overall accuracy occurred between the first and second phases, namely form the Maximum Likelihood classifier (North image 73%, South image 76%) to around 80% in the final mapping. |
|---|