A carregar...
Publicação
Fusão de imagens de satélite utilizando informação de invariantes e séries temporais
| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 10.26 MB | Adobe PDF |
Autores
Orientador(es)
Resumo(s)
Para realizar a monitorização da superfície terrestre, diversos satélites e constelações de
satélites orbitam a Terra. Diferentes satélites oferecem diferentes resoluções electromag-
néticas, diferentes resoluções espaciais e diferentes resoluções temporais.
Os algoritmos de fusão de imagens de satélite, combinam satélites de maior resolu-
ção espacial e satélites com uma maior resolução temporal. Estes algoritmos permitem
estimar imagens de satélites à melhor resolução espacial possível e com maior frequência,
resolvendo problemas parciais de qualidade das imagens originais, e.g. contaminação por
nuvens. As soluções existentes de fusão criam estimativas bastante exatas das zonas de
observação, no entanto ainda têm problemas derivados das zonas heterogéneas. Tendo em
conta que nenhuma das soluções analisadas utiliza informação adicional para melhorar
os resultados das fusões, existe potencial exploratório.
Neste documento foi proposto um novo algoritmo de fusão de imagens de satélite
denominado LCBFM. Utiliza uma segmentação que agrupa os píxeis de acordo com o
seu comportamento espectral temporal. Este algoritmo foi criado com base em dois estu-
dos realizados ao longo desta tese. O primeiro destes consiste na análise do impacto de
invariantes sobre os resultados da fusão de imagem. O segundo foi acerca de segmenta-
ção temporal aplicada ao agrupamento de píxeis pelos seus comportamentos espectrais,
incluindo uma análise de diferentes tipos de algoritmos de segmentação e distâncias.
Para validar o nosso algoritmo, as imagens por este estimadas foram comparadas com
as de outros algoritmos conhecidos na literatura, o STARFM, ESTARFM, STAIR e FSDAF.
Para além destes foi comparado também com o ESTAIR, um algoritmo desenvolvido num
trabalho anterior que obteve melhores resultados que os antecessores em que se baseou,
e serviu como base para o nosso algoritmo. O LCBFM obtém resultados bastante bons e
particularmente interessantes para regiões heterogéneas.
To monitor Earth’s surface, several satellites and constelations of satellites orbit around the planet. Different satellites offer diferent electromagnetic resolutions, diferent spatial resolutions, and different temporal resolutions. The satellite image fusion algorithms combine higher spatial resolution satellites with a higher temporal resolution. These algorithms allow for an estimate of these images at the best possible spatial resolution and highest frequency making them less prone to existing quality problems in the original images, e.g. cloud contamination. The existing solutions for fusion generate accurate estimates of the regions of interest, although, there are still problems when it comes to heterogeneous areas. Considering that none of the analized solutions uses aditional information to better their results, there seems to be a potential for improvement within this area. In this document, we propose a new satellite image fusion algorithm named LCBFM that uses a clustering that groups pixels accordingly to their spectral behaviour in time. This algorithm was created based on studies developed throughout this thesis. The first of these consists in the analysis of the effects that invariants have on the results of image fusion. The second was on temporal clustering applied to grouping pixels with similar spectral behaviours, which includes an analysis on several types of clustering algorithms and distance measurements. To validate our algorithm, the estimated images were compared to those calculated by other known algorithms in the literature, such as STARFM, ESTARFM, STAIR and FSDAF. Along with these, ESTAIR was also used as a comparison basis, which was developed in a prior work, obtaining great success and served as the basis for our algorithm. LCBFM reaches great results, with these being particularly interesting in heterogeneous regions.
To monitor Earth’s surface, several satellites and constelations of satellites orbit around the planet. Different satellites offer diferent electromagnetic resolutions, diferent spatial resolutions, and different temporal resolutions. The satellite image fusion algorithms combine higher spatial resolution satellites with a higher temporal resolution. These algorithms allow for an estimate of these images at the best possible spatial resolution and highest frequency making them less prone to existing quality problems in the original images, e.g. cloud contamination. The existing solutions for fusion generate accurate estimates of the regions of interest, although, there are still problems when it comes to heterogeneous areas. Considering that none of the analized solutions uses aditional information to better their results, there seems to be a potential for improvement within this area. In this document, we propose a new satellite image fusion algorithm named LCBFM that uses a clustering that groups pixels accordingly to their spectral behaviour in time. This algorithm was created based on studies developed throughout this thesis. The first of these consists in the analysis of the effects that invariants have on the results of image fusion. The second was on temporal clustering applied to grouping pixels with similar spectral behaviours, which includes an analysis on several types of clustering algorithms and distance measurements. To validate our algorithm, the estimated images were compared to those calculated by other known algorithms in the literature, such as STARFM, ESTARFM, STAIR and FSDAF. Along with these, ESTAIR was also used as a comparison basis, which was developed in a prior work, obtaining great success and served as the basis for our algorithm. LCBFM reaches great results, with these being particularly interesting in heterogeneous regions.
Descrição
Palavras-chave
Fusão de imagens Moderate Resolution Imaging Spectroradiometer(MODIS) Sentinel Landsat Reflectância da superficie Deteção remota