Interpretação de imagens multitemporais de sensoriamento remoto.

• Autor(a) principal: Alves, Andrei Olak
• Data de Publicação: 2011
• Tipo de documento: Dissertação
• Idioma: por
• Título da fonte: Biblioteca Digital de Teses e Dissertações da UERJ
• Texto Completo: http://www.bdtd.uerj.br/handle/1/10816
• topic: Remote sensing; Interpretation of multitemporal images; Fuzzy Markov chains; Automatic supervised classification; Sensoriamento remoto; Interpretação de imagens multitemporais; Cadeias de Markov Fuzzy; Classificação automática supervisionada; CNPQ::ENGENHARIAS

Multitemporal remote sensing data analysis is concerned to the thematic mapping of land use in different instances of time, aiming to identify changes in land cover/use in such period. Most multitemporal studies in the literature, while producing the thematic mapping, merely perform separate supervised automatic classifications of the remote sensing images. That is the so-called monotemporal approach, in which the classification process do not take advantage of the temporal dynamics knowledge. Recent essays conducted over the past decade had indicated that the use of a model of knowledge about the dynamics of the region (temporal transformation model) based on Fuzzy Markov Chains (FMC) provides superior results than the supervised monotemporal classifiers. This master's dissertation focuses on one aspect of such approach still not enough investigated: the combination of FMC for short time intervals to classify images of longer periods. Herein, the study area is a forest remnant in the municipality of Londrina, Paraná, in south of Brazil, which encompasses the borders of the Mata dos Godoy State Park. The database is composed by five Landsat 5 TM images, each one acquired five years after the previous. The experiments showed that the use of Fuzzy Markov Chains to model the temporal dynamics can improve the performance of the classification process, while compared to the conventional monotemporal approach. Besides, it could be observed that the more recent is the past information the better is the classification performance. Other important aspect that could be noticed was the reduction of the superiority of direct estimation of FMC over its extrapolation, as well temporal distance increases. This approach can contribute to the increment of the degree of automation of multitemporal images sequence classification. Considering the improvement in the classification performance, its application could also speed up the land use / cover monitoring.