A maximum likelihood approach for detecting a step pattern of change and locating the transition instant in a sequence of multitemporal SAR images

In this work we derive two different detectors that fully exploit the information available in a multitemporal sequence of M SAR images with regions possibly affected by step changes in reflectivity, with unknown transition instant. These detectors process the whole sequence using a maximum likelihood approach to: (i) locate the transition instant, and (ii) discriminate between regions affected and not affected by a step change. The KSP-detector assumes that the two mean RCS levels are known, while the USP-detector assumes that the two mean RCS levels are unknown. Two different approximations for the probability of correctly locating the transition instant for the KSP-detector are considered, thus obtaining two analytical performance expressions. The practical effectiveness of the technique is shown by applying the USP-detector to a sequence of 10 ERS-1 SAR images of forest and agricultural areas