inside view

Our work focuses on commercial vehicle classification using what is called attributed SAR imagery. In attributed imagery, each pixel is assigned properties beyond intensity and relative coordinate to aide an object classification recognition. The current work on pixel selection offers insight into how to select pixels by type of attribute so as to give insight into the performance of a classification algorithm. Our approach, using the spectrum parted linked image test (SPLIT) algorithm, does not require processing other than image formation, but SPLIT also provides only a partial set of attributes. Hence, the trade-off is simplicity and efficiency versus fidelity. It is possible, therefore, to use SPLIT with careful local peak detection to select pixels, attribute those pixels, and then seed higher fidelity algorithms to gain efficiency in those algorithms.