Research And Application On WNN Based On Multi-subgroup Structure

Particle Swarm Optimization algorithm （PSO），as a member of the intelligencecommunity, is put forward to by the simulation of intelligent phenomenon expressed throughbiological communities behavior on prey. Wavelet neural Network, which is combined withWavelet transformation theory and artificial neural Network, is a new kind of feedforwardneural Network model. It fully inherits good time-frequency localization of Wavelettransform and the self-learning function of the neural Network. It has the strong capability offunction approximation and fault-tolerant ability.The PSO algorithm is used to train the parameters of the WNN instead of the traditionalgradient descent algorithms, and the threshold and connection power value of the neuralnetwork are set as the particle position vector of the PSO algorithm. The optimal parametersvalue of neural network is searched by updating the velocity and position vectors of theparticles The generality of the training program is improved by using PSO algorithm to trainthe parameters of WNN. The PSO algorithm avoids the blind search of the network, andreduces the iteration times. To a certain extent, it makes the network easily out of the localminimum value of the problem.In this paper a new algorithm is presented, which is based on wavelet neural networkand multiple sub-particles. The particle swarm is divided into two different sub-groups, andthey take respective evolution strategies. The two subgroups achieve the purpose ofoptimization through competitions in the group and between groups. The WNN is put intoimage denoising field, with improved particle swarm optimization algorithm to optimize thenetwork. A kind of image denoising algorithm is presented, which is based on the medianfiltering method. Simulation experiments show that the algorithm can achieve good imagedenoising and protect the detail of the image information to the maximum extent.