Research On A Neural Network Pattern Recognition Method Based On Optimal Classification Face And Applications

Neural network method has generalization capability, and is widely used in pattern recognition. Generalization means: trained with a training examples set, the network can also recognize examples never met before.Though the method has generalization capability, it doesn't satisfy the application. When the examples include yawp, wrong recognition occurs. Generalization of the traditional is not ideality and instability. The instability means: for the same task, when training examples change, the generalization capability also changes. In this paper , we present a new approach of improving the generalization and the stability of neural networks. Proved by the XOR and double helices problems, the new approach is effective and satisfying.Using AB neural network model, the convergence is speeded up. When the classification face is complicated, we cut it into many little faces. And one as A network, the others as B networks. The training time is only 2/3 of before.Email is widely used, and the problem of spam mails is more and more serious. Using the method presented in this paper, the recognition rate of the Email filtering is greatly improved.