A New Parameter Selection Method of Neural Network

Abstract

A new parameter selection method of neural network is presented after researching parameter selection method of neural network deeply, and it is applied to surface defect online inspection system of cold rolled strips. The method exerted the merits of small-samples fully which was utilized to train and test neural network by every possible combination of parameters to get a group of neural network parameters by plotting histogram of recognition rate under different parameter combinations, and the parameters are regarded as optimized parameters of neural network. Experiments showed that a best recognition effect by using the parameters for neural network which are selected by the new parameter selection method can be achieved among all the parameters selected randomly for surface defect of cold rolled strips.