Efficiently Computing and/or Compensating for Object Variability for Automatic Target Recognition (ATR) Applications

Variability exists in all objects considered in ATR data sets. The two most important sources of this variability are (1) added or removed objects (example: object attached to the back of a tank) or (2) a part displaced relative to other parts, sometimes referred to as “articulation” (the classic example is a tank’s moveable turret and gun). These can cause severe difficulties in comparing measured data from an operational system with computed data sets. The turret example provides the case in point: there are essentially an infinite number of combinations of turret rotations and gun angles; computed data sets to cover all angles and frequencies are simply out of the question. As the SBIR description outlines, the two approaches are to simplify the data set (somehow) or to alter the ATR algorithm itself to ignore or compensate for the variability. This SBIR program will examine both aspects, although the main focus will be on examination of data sets and computational techniques with an eye on simplifying the data generation process by ignoring all but the essential elements of the scattering to examine errors from “full” signatures.