Estudo Comparativo de Alguns Classificadores Utilizando-Se Imagens RADARSAT DA Regiao de Tapajos

The digital classification is a very important procedure used in remote sensing applications. Several classification techniques had been developed and tested for optical data. With respect to radar data, the development, evaluation and testing of digital classifiers have been a subject of study in last decades. The objective of this article is to compare the classification results of a SAR image, obtained using three diferent supervised classifiers. A RADARSAT image from the Tapaj6s National Forest was used for this purpose. The original image was processed in one-look amplitude. This image was degraded obtaining a nine nominal look image. The degraded image was filtered by the Frost filter in order to analyse the influence of the speckle reduction on the classification. The effect of textural information was analysed by filtering the image by the sample coefficient of variation and by an estimated parameter of the amplitude κ distribution. The classifiers used were the Maximum Likelihood, Iterated Condicional Modes, and a region classifier based on the Bhattacharrya distance. The image was classified in three classes: Bare Soil and Pasture, Regeneration (Secondary Forest), and Primary Forest. The classification was evaluated by the Kappa coefficient of agreement and by the corresponding confusion matrices, using samples obtained from a field work.