Research On BP Network Classification And Classification Capabilities Hopfield Network Based Association

This paper mainly studies the two typical artificial neural networks inthe artificial neural network--BP network and Hopfield network. We deeply study theClassification and recognition of the two networks, and compare the advantages anddisadvantages of the two networks. Then we improve the recognition of BP networkin handwriting. And to study the characteristics of handwritten image extraction, wetest its effect by BP network, and the test results are satisfied.Neural network can be used in pattern recognition. This function has great utilityvalue. This paper studies the recognition ability of BP network and the discreteHopfield neural network. It compares the identification ability of BP network anddiscrete Hopfield network in numbers, letters, handwritten digit, receiving theconsiderable conclusion that the noise in a certain range, the recognition effect ofdiscrete Hopfield neural network is better; when the noise exceeds this range, theerror will be increased rapidly, in this range, the BP network identification result isbetter.In this paper, we talk about the issue of handwritten numeral recognition, andproposes an identification method based on BP neural network. What we do is tochange the location and size in the picture. We found that, they both directly affectpicture recognition effect. This paper is firstly extracted the contour of the number,and then normalize them into the picture of. To do so, not only makes the imageregion the same size, but also in the image center. This makes the recognition moreideal, to achieve high identification purposes. In addition, we choose BP network withbetter tolerance, and200groups of handwritten digital image as the input vector,110groups of the other as recognition, the efficiency reached more than90%.Also，we talk about another improvement about the handwriting recognition ofBP. Network. With the development of computer network, more and more electronicproducts use handwriting input. Therefore, the identification of these handwrittenimages becomes increasingly important. This paper is using the network modelwith one hidden layer. Feature extraction refers to using computers to extract theimage information, to decide whether each image point belongs to an imagefeature. The results of feature extraction are to separate those images into differentsubsets, these subsets tend to the isolated point or continuous curve or continuousarea. In this paper, we extract7invariant moments (Hu moments), mean, variance andthe characters of the image potential, with these characters to identify the images. The recognition accuracy is above80%.