Detector robustness to change in depression angle

Many ultra-wideband (UWB) synthetic aperture radar (SAR) detection algorithms employ some combination of a set of features, calculated from the incoming raw radar data return, to segregate targets from clutter in a SAR image. Based on the training data, the algorithm designer selects those features that exploit some difference in the physical characteristics between the target class and clutter class. A detection algorithm is then trained to determine values for a set of algorithm parameters that will minimize some sort of error criterion. The physical characteristics that guide the feature selection can change, however, with changes in the attributes of the data collection, such as the depression angle from the radar to the point of interest. When the depression angle changes, the algorithm parameters that were optimal for the training data may no longer be optimal for test data at a different depression angle. We examine the changes in detector performance resulting from depression angle mismatches between the training and test data sets.