The effect of SAR imaging conditions on classification algorithms

Synthetic Aperture Radar (SAR) is used for a variety of missions and many of these missions require higher resolutions, larger scene sizes, and a variety of platform flight paths. All these factors affect the signature of the vehicles and cause the signature of the same vehicle to look significantly different even if the vehicle is at the same azimuth and elevation angle with respect to the center of the synthetic aperture, and even if the radar has the same bandwidth and coherent integration angle. Vehicle signatures are compared both qualitatively and quantitatively using long integration times, different flight paths, and different image locations with respect to scene center. The images are quantitatively compared using a metric based on previously successful SAR classification approaches. It is shown that the classifier must account for these signature differences to achieve reasonable performance.