Mixture model for the segmentation of the InSAR coherence map

Abstract

In this work, we classify the interferometric SAR (InSAR) coherence map into three classes using the Bayes' theorem. The segmentation procedure is performed using a mixture modelling of the coherence map. The multimodal density of the mixture comprises three component functions characterizing different land surface categories (lake, bare soil, urban ...). This work is an ameliorated segmentation approach of that published by the authors in R. Abdelfattah, et. al., (2006). We test the performance of the proposed mixture model on a dataset about regions with different geophysical characteristics and different time interval between the acquisitions. The results of this study could be used as a supervised learning step for an automatic land cover classification algorithm. This new method classifying the image considering the corresponding InSAR coherence map is particularly powerful for the detection of layover and shadow regions.