False alarm mitigation techniques for SAR CCD

Coherent Change Detection (CCD) is a powerful technique for detecting fine scene changes between two Synthetic Aperture Radar (SAR) images taken at different times. SAR CCD imagery can detect ground disturbances that are invisible in optical or traditional SAR imagery, such as footprints or vehicle tracks [1]. One particular problem with the extreme sensitivity of CCD is the presence of false alarms (clutter) introduced by phenomena such as low SNR (esp. radar shadows) and vegetation [2]. We present two methods to improve the sensitivity of the detector while reducing the amount of false-alarms. The first uses a generalized likelihood ratio test for change detection which incorporates noise explicitly in its models. The second combines two CCD images, generated from three SAR passes of the same area, to cancel out false alarm regions and show only changes from man-made activities of interest, such as vehicle tracks. We show results from each algorithm on real data, and find that the algorithms are effective at reducing the amount of false alarms while increasing the sensitivity of our detector.