Genetically Optimised target detection

Traditional techniques for identifying targets in clutter are not necessarily robust with respect to orientation of the target, partial obscuration, variations in lighting, etc.. Neither are they necessarily computationally undemanding.

Genetic Programming is a technique that builds programs by natural selection. The technique requires no prior knowledge other than the ability to quantify the effectiveness of the candidate programs at the task in hand.

The measure of effectiveness, or fitness, can incorporate a measure of the compactness of the program. The genetic programming technique will then evolve small programs that are effective at the given task.

In this application we are interested in building primitive target detectors analogous to the matched filtering that occurs in the mammalian retina and layers of the visual cortex.

Genetic Programming produced small programs that matched particular targets irrespective of their position, orientation, illumination or magnification. These small programs were computationally undemanding to the extent that they could be easily built in hardware for real-time operation.