A theoretical analysis of a ranking discrimination algorithm

The authors summarize the Lincoln Laboratory multistage target detection algorithm for high-resolution synthetic aperture radar (SAR) imagery. The second stage of the algorithm, called the discriminator, consists of a one-class quadratic discriminant computed from various target features. The output of this algorithm is a list of potential targets, which can be ranked according to their `targetness´ before being fed to the final classification algorithm. The ranking score can be used for prioritization in a real-time system. The authors theoretically analyze the ranking discrimination algorithm and compute performance examples using representative parameter values. The theoretical expressions can be used as a guide to help allocate computer resources in the design of a real-time, wide area search high resolution SAR system