A generalized likelihood ratio test for SAR CCD

Coherent change detection (CCD) is a powerful technique for detecting minute disturbances in synthetic aperture radar (SAR) imagery. Coherent change detection uses a test statistic to compare, pixel by pixel, two or more SAR images of the same scene. Coherent change detection can detect very small disturbances, not normally visible in SAR or optical imagery, such as footprints or vehicle tracks. The literature describes a number of different detection statistics, the choice which will effect both the contrast of the detected disturbances and the amount of false or uninteresting detections. We present a generalized likelihood ratio test for change detection. Our effort improves upon previous work by incorporating noise in our models, and by optimizing the likelihood parameters separately at each pixel. We compare results from the GLRT with the standard coherence metric on a number of different examples of collected synthetic aperture radar data. The results show that the GLRT provides a useful improvement to the CCD processing.