The Study of Improved CP Neural Network for Remote Sensing Classification
重复传播网络的改进及其在遥感分类中的应用研究

The study improves the CP neural network including the weight vectors initial, network learning, the competitive nodes dynamic allocation and the vigilance parameter design etc. It applied to the remote sensing imagery classification. The learning times influence the node numbers and learning accuracy. According its raise, the learning accuracy increases a limit degree.The research boundary is decided by the vigilance parameter , which decreases with the learning process in order to get the high accuracy and the stability of networks. The two\|layer networks improves the overall accuracy by 1.1%, Kappa coefficient by 0.02, which compared to the single\|layer networks. It is concluded the improved CP is another classification method for remote sensing.