Automatic target recognition (ATR) and classification is a computationally demanding task, but with the recent increase in the computing power for industry standard FPGAs and DSPs it has become a feasible and very useful application in military sensing equipment. The ATR method presented here uses Zernike moments of binary representations of infra-red targets for the classification process. Zernike moments are known for their good image representation capabilities based on their orthogonality property. They are often used because the magnitude of the moments provides rotation and scale invariance. For the detection of the target candidates, a given region of interest (ROI) is searched for possible target signatures using a simple threshold segmentation. From the resulting binary objects, the biggest or center-most object can be selected. For this target, the minimum enclosing circle is determined using the bounding box found during the segmentation process. This minimum enclosing circle is scaled to the complex unit disk, where Zernike moments are defined. The moments up to order five are then computed directly from the binary image using a fast recursive algorithm. The resulting twelve-dimensional moment magnitude vector is then classified with a 1-NN algorithm, where a set of class templates has been pre-computed off-line for each class using a simulated annealing approach for cluster analysis.