The Research On Neural Network With Fourier Weight Function And Its Application In Image Recognition

After years of research, the achievements of artificial neural network are plentiful. However, the weights of the traditional neural networks(BP,RBF) are constant, and the training weights could not reflect the information of the sample.In additional, the model of traditional neural networks is difficult to determine and the number of hidden layers needs to be tried and modified. Document one and document two proposed a new neural network algorithm which is spline function weight neural network training algorithm. This training algorithm overcame the defects such as partial minimum frequently, low convergence speed, sensitive to initial value which exists in traditional neural network,which will lead neural network to a new phase.Based on spline weight function neural network, Fourier weight function neural network has been proposed in this dissertation. In theoretical section,firstly,the model of Fourier weight function neural network has been constructed. secondly,the solution of Fourier weight has been given. third,we get the expression of network error combine with Fourier weight function neural network. At last,through MATLAB simulation results,the dissertation proved Fourier weight function neural network has high approximation accuracy, fast training speed and strong generation ability compare with the traditional neural networks.In this dissertation, the Fourier weight function neural network has been used in texture image recognition. Image recognition is a typical pattern recognition problem. Four sets of texture parameters has been extracted according to the analysis of grey level co-occurrence matrix, which are used for textural image classification and recognition as the input of Fourier weight function neural network. The MATLAB simulation results show that Fourier weight function has better precision in recognition compared with BP and RBF neural networks.