Detecção de mudanças suave a partir de imagens de fração

Automatic change detection is an important challenge in multitemporal remote sensing applications. Some studies have been demonstrated that fraction images can be useful for land cover change detection because the radiometric problems are minimized and the interpretation of changes in land cover is facilitated, since the fractions have a physical meaning. Furthermore, interpretations at the subpixel level are possible. Because the sampling training is not required, unsupervised techniques are often chosen for the generation of change maps without intervention of the analyst. This work presents a soft classification procedure for detecting changes between a pair of fraction images. The technique is applied after the generation of binary change maps by a hard classifier based on Mahalanobis distance. The hard algorithm requires multivariate normality assumption for the differences among fractions, but the soft one logistic regression assumes nothing about the probability distribution of variables, requiring the estimation of few parameters. The soft classification allows estimating the probability of each pixel belonging to the change class; these probabilities are used to create a gradual change map. This approach was tested using a pair of synthetic images and the results were analyzed in terms of Mean Square Error (MSE). The results indicate that the soft classification have produced better results than only using the hard algorithm.Pages: 8122-812