Evaluation of two region based classifications in Tapajós National Forest using the ALOS/PALSAR polarimetric and interferometric coherences

The use of phase information present in complex multi polarized images may increase the classification results. Thus, the coherence is one attribute that may be extracted from these images and used to distinguish some land cover classes. Therefore, its discriminatory capability for land use and land cover classification is analyzed. The analysis is based on the classification results of a region classifier, which needs a segmented image as one input. The influence of this kind of image input is also evaluated using of two segmentation algorithms, the SegSAR and the SPRING region growing. Two ALOS/PALSAR images acquired over Tapajós National Forest in the Brazilian Amazon were classified. The classifications were quantified by the overall accuracy, the kappa values and its variance. The classification improvement using the coherence information with intensity images was noticed for every image set.