MIMOSA: An Automatic Change Detection Method for SAR Time Series

This paper presents a new automatic change detection technique for synthetic aperture radar (SAR) time series, i.e., Method for generalIzed Means Ordered Series Analysis (MIMOSA). The method compares only two different temporal means between the amplitude images, whatever the length of the time series. The method involves three different steps: 1) estimation of the amplitude distribution parameters over the images; 2) computation of the theoretical joint probability density function between the two temporal means; and 3) automatic thresholding according to a given false alarm rate, which is the only change detection parameter. The procedure is executed with a very low computational cost and does not require any spatial speckle filtering. Indeed, the full image resolution is used. Due to the temporal means, the data volume to process is reduced, which is very helpful. Moreover, the two means can be simply updated using the new incoming images only. Thus, the full time series is not processed again. Change detection results between image pairs are presented with the airborne sensor CARABAS-II, using a public data release, and with TerraSAR-X data. In the case of time series, change detection results are illustrated using a TerraSAR-X time series. In every case, the MIMOSA method produces very good results.