Abstract:
The traditional adaptive resonance theory 1(ART1) neural network is widely used in pattern recognition,but the classified result of ART1 is sensitive to the sequence of the input patterns.If the sequence changed,the result changed too.Its limitation is analyzed and the cause of it is that traditional ART1 neural network ignore the effect of each input pattern’s zero element on the result of classify.A new improved ART1 neural network is proposed and it can solve the problem mentioned above.Experiment results show the improved ART1 neural network is effective on English characters recognition.