Comparison between spatial and temporal estimation of entropy on polarimetric SAR images

Polarimetric entropy is used in various classification algorithms, often based on the H/A/α feature space. Entropy, as all the statistical parameters, has to be estimated. This is often made through a boxcar estimation of the covariance matrix. In this process, different pixel populations are mixed in urban areas. The estimated entropy tends to show a poor contrast that decreases the classification results. We propose to estimate entropy using temporal series. This temporal estimation of entropy is compared to the classical spatial estimation over urban and natural areas. Temporal entropy highlights areas that are stable in time. This permits to discriminate targets that are not be detected using spatial entropy.