Hyperspectral change detection by direction pursuit

Change detection is an important area of interest within the hyperspectral community. Generally, a first step in the detection consists in predicting some general changes as shadows or atmosphere evolution which should not be detected, and in a second step the local changes are detected. Here we choose the covariance equalization to predict those changes. We present in this paper a change detection method based on an anomaly component pursuit algorithm, namely ACP, recently proposed for anomaly detection, which combines anomaly classification and detection. We experimentally show the efficiency of this method, and we compare the results obtained with those of the classical RX detector. We also compare this pursuitbased change detector to two information-based change detection methods, using respectively the Kullback-Liebler divergence information, and the Kendall´s tau dependence measure. We show that in our simulation conditions, the ACP algorithm gives very interesting results for change detection analysis.