Evolve-Based Cloning Template Design For CNN

Cellular Neural Network (CNN) template parameters play a important role on CNN’sdynamic properties, determines the ability of CNN in terms of Memory and correct recall thenoise interference data. The current template design methods are mainly based on the analysismethod, local learning method and global learning method. Based on the analysis of thetraditional methods and local learning method without considering the template robustness,but the global learning method such as genetic algorithm (GA) and particle swarmoptimization algorithm (PSO), and other heuristic optimization algorithm can effectivelysolve this problem.Associative Memories Cellular Neural Networks (AMCNN) use traditional particleswarm optimization (PSO) algorithm evaluation function without considering the templaterobustness. In this paper, based on three evaluation criteria PSO for Cellular NeuralNetwork template design method (MEPSOCNN) make particle swarm evolution moreambitious, quickly find the template which accord with robustness requirements and havegood memory ability. The first layer of the evaluation criteria to ensure the CNN can besuccessfully stored all the original mode; The second layer evaluation criteria CNN has goodnoise immunity; third layer evaluation standard make particle swarm faster to find a templaterobustness requirements solution.The experimental results indicate that the proposed PSO base on three evaluation criteriaapplied to CNN design(Multiple Evaluate Particle Swarm Optimization Cellular NeuralNetworks-MEPSOCNN) in most experimental results better than the PSO based on singleevaluation criteria (Particle Swarm Optimization Cellular Neural Networks–PSOCNN).Conclude finally, when the associative memory cellular neural network input patterns are lownoise, MEPSOCNN is a stable and efficient space-invariant cloning templates design method.