NONLINEAR PATTERN CLASSIFICATION ASSOCIATED WITH CELLULAR NEURAL NETWORKS-BASED DYNAMIC PROGRAMMING

A Cellular Neural Networks (CNN)-based nonlinear pattern classification algorithm utilizing the most likely path-finding feature of dynamic programming is proposed. Dynamic programming for the most likely path-finding algorithm can be implemented with CNN. If exemplars and test patterns are assigned as the goals and the start positions, respectively for our CNN-based dynamic programming, the paths from the test patterns to their closest exemplars are found with the optimality feature of CNN-based dynamic programming. Such paths are utilized as aggregating keys for classification. Our algorithm is suitable for patterns with nonlinear pattern boundaries. Simulation results are included.