Research On Visual Image Classification Method Based On Spiking Neural Network

In recent years,the Convolutional Neural Network(CNN)has solved the problems of computer vision,natural language processing and pattern recognition in the field of deep learning,and it is widely used.The model of convolutional neural network is simple to train,but it has the problem of high energy consumption and high resource demand in application.The Spiking Neural Network(SNN)is more efficient in simulating biological nerves.Compared with convolutional neural networks,it has high performance characteristics,but the training of this network model is difficult.This paper proposes a method to map a convolutional neural network into a spiking neural network.The basic idea is to combine the convolutional neural network with the spiking neural network and apply the weighted parameters of the convolutional neural network model to the spiking neural network.Therefore,the problem of difficult training of pulse neural network is skipped,and the simulation on the pulse neural network is realized.The experimental results show that the proposed method can make the accuracy of image classification in spiking neural network close to that of convolutional neural network.The main work of this paper is as follows:Firstly,the convolutional neural network model with image classification function is constructed.The combination of the number of convolution kernels and the pooling layer in the model is set differently and applied to the image classification problem.The experiment analyzes the classification accuracy and obtains reasonable parameter settings.Secondly,the visual processing(image classification)process simulation of spiking neural network is proposed by convolutional neural network.Through the study of the neural network and network structure of the spiking neural network,it can be found that the spiking neural network can realize the function of the network by means of the convolutional neural network.The simulation method solves the problem of difficult training of the spiking neural network and can achieve the accuracy close to the convolutional neural network.The spiking neural network model obtained by mapping is compared with the classification accuracy and power consumption of the convolutional neural network.The results show that the accuracy of the former is slightly lower than that of the latter,but the advantage of spiking neural network is obvious in power consumption,and this is acceptable.Thirdly,by referring to the idea of feedback adjustment algorithm(BP)in convolutional neural network training,a threshold setting tuning algorithm based on feedback adjustment is proposed to optimize the threshold of neuron groups in spiking neural networks.After tuning,the classification accuracy of spiking neural networks is improved.