Handwritten Figure Fuzzy Recognition Based On Particle Swarm Neural Network

Over the years, handwritten figure recognition has been one of the central issues on research of pattern recognition, handwritten figure recognition has a wide range of applications in a particular environment, such as automatic recognition system of postcode, tax returns and checks automatically treatment systems of bank and so on. Although the figure only has 10 kinds of categories, and the stroke is simple, its recognition problem does not seem very difficult. However, the key technology of recognition has not been resolved, therefore, the current recognition system still hardly to achieve the high recognition accuracy.In this paper, a handwritten figure is take for the research object, first of all the handwritten figure is pretreated, such as gray of image, binarization, denoising, tilt adjustment, normalization, thinning and so on, eigenvector for algorithm training is extracted based on the combining method of statistical characteristics of partial Fourier transform and wavelet transform, Radon transform, Legendre moment features, extraction of pen road density function and the extraction method of the micro-structural characteristic in structural characteristics. In the implementation of algorithm, the application has a good fault-tolerant capability, sorting capability, parallel processing and self-learning ability on artificial neural network, by improving the transfer function, mediating of study rate and adding noise training in neural network, the stable network which is easy to convergence is obtained, in order to overcome the exist problem of neural network training algorithms in handwritten figure recognition easily fall into local minimum value, this paper presents that the particle swarm optimization algorithm training neural network of adaptive inertia weight is adopted, namely that using the optimal weights and thresholds of particle update iterative training neural network, in which the inertia weight of update particle is improved self-adaptively, and finally recognition output of handwritten figure is implemented by using fuzzy recognition.A secondary neural network used for handwritten figure recognition is set up in this paper, the simulated experiment is implemented at Matlab7.0 environment. With the comparison on two handwritten figure recognition transfer function and mediation to improve the learning rate and add noise to deal with the neural networks and only particle swarm optimization neural network, the result shows that handwritten figure fuzzy recognition algorithm of particle swarm neural network can effectively optimize the neural network to avoid "precocious" phenomenon of network, which greatly raised the accuracy of network training, accuracy rate of handwritten figure recognition for network has been improved obviously.