Evolving finite state machines with embedded genetic programming for automatic target detection

Abstract

This paper presents a model comprising Finite State Machines (FSMs) with embedded Genetic Programs (GPs) which co-evolve to perform the task of Automatic Target Detection (ATD). The fusion of an FSM and GPs allows for a control structure (main program), the FSM, and sub-programs, the GPs, to co-evolve in a symbiotic relationship. The GP outputs along with the FSM state transition levels are used to construct confidence intervals that enable each pixel within the image to be classified as either target or non-target, or to cause a state transition to take place and further analysis of the pixel to be performed. The algorithms produced using this method consist of nominally four GPs, with a typical node cardinality of less than ten, that are executed in an order dictated by the FSM. The results of the experimentation performed are compared to those obtained in two independent studies of the same problem using Kohonen neural networks and a two stage genetic programming strategy.