Application Research On Shape Recognition System Based On Associative Memory Neural Network

Shape recognition is an important research area of pattern recognition. With the research development of associative memory neural network,now the research hot point is the application problem of associative memory neural network. In this paper,we discussed some key application problems of shape recognition system based on associative memory neural network.The first problem is that how to encode the input pattern. As associative memory neural network require the input signal must be binary code and the character vectors are always real number,we must encode the input signal to be binary code. The second problem is that associative memory neural network has a dynamical course,when the network fall into stable state,it will be the state we wanted or be the false state. Then the problem is how to express the mean of the stable state of the network explicitly. The third problem is how to choose learning samples to improve the rate of recognition.For the first problem,we overcome the limitation of binary code algorithm based on threshold value,which is popular used now,we presented the binary code algorithm based on the statistical information of neighbour region. For the second,we presented the associative memory neural network output discriminant algorithm based on pattern similarity to solve it. For the third problem,based on the work above,we presented the samples choosing solution,which is based on clustering and multi-category.With the experiment result,we can see these methods presented above are effective to solve the problems presented. Solving these problems does very much good to the practical application of Shape recognition system.To the practical problems,we found that the result is not very good when we only utilize pattern's one character. To the unrecognized shape,we can get three characters:the numbers of peak,the first station of peak,and the distance of two peaks. So,the fourth problem is how to synthetically utilize pattern's characters to shape recognize in order to improve the rate of recognition. To do it better,we first analyze the degree that pattern's characters effect the result of shape recognition. Then we presented the shape recognition system based on information fusion technology and we implemented the system primarily.