Abstract: Based on the Hebb learning rule, noised and distortion figures of 0~9 were identified, using the associative memory ability of discrete Hopfield neural network. Through improving the memory samples, which is to be orthogonal, and using Hebb rule to learn the improved memory sample, the weight value matrix was obtained, the noise figure would be identified according to the information of noise figure. The identification experiment on noise figure by using the improved Hopfield neural network shows that the method improves the memory capability and the correct identificanon rate of traditional network.

Key words: Discrete Hopfield, Orthogonalization, Noise-figure recognition