Research On Image Recognition Algorithms On Deep Learning

In recent years,with the rapid development of Internet technology,image data not only increases exponentially,but also becomes more diverse.Traditional recognition methods can not meet people's needs.Deep learning based on bionics theory emerges as the times require.By simulating human brain,it can automatically learn and extract features,giving full play to the advantages of big data.It is of practical significance to apply depth learning to image recognition and improve the efficiency and accuracy of image recognition.Aiming at the problems of low recognition accuracy and low recognition efficiency of traditional multi-layer perceptron model in handwritten numeral recognition,an improved multi-layer perceptron model is proposed.Dropout is introduced to solve the over-fitting problem,Adagrad optimizes the parameter debugging process,and ReLU solves the gradient dispersion problem.The experimental results show that the improved model can effectively carry out feature learning and perform well in improving recognition performance.Aiming at the problem that the soft Max classifier method used in traditional convolution neural network can not effectively take into account the recognition accuracy and recognition efficiency,this paper proposes the method of using Parzen classifier combined with convolution neural network for image recognition.In feature extraction,the convolution neural network is used to extract image features;in classification and recognition,drawing on the advantages of traditional classifier,Parzen classification is proposed.Apparatus method is used for image recognition.The experimental results show that Parzen classifier can effectively classify images,and the algorithm has good generalization performance.In order to further improve the recognition accuracy,convergence and generalization performance of image recognition algorithm,a multi-layer perceptron image recognition algorithm based on vector neuron is proposed.Firstly,the length of the active vector is used to represent the probability of entity existence and the direction is used to represent the instantiation parameters.Then,an iterative successive routing mechanism is used to update the parameters.The experimental results show that the improved algorithm performs well in recognition efficiency and accuracy,and has good convergence and generalization.