The Research And Verification Of Image Classification Based On Convolutional Neural Network Optimization Algorithm

Convolutional neural network is a kind of efficient recognition method.Convolutional neural network has three characteristics,which are parameter sharing,local perception and subsampling respectively.These characteristics make the training parameters decrease and the speed of the train increase,making the train in the training process have a good performance.Currently,convolution of the neural network has been utilized in all aspects of life well,especially in the image classification tasks,voice recognition,road sign recognition and so on.But there are still some problems in its development process.In this paper,the convolutional neural network in the field of image classification are researched.The purpose is to improve the accuracy of image classification and reduce the error rate.The activation function preserves and maps the characteristics of the activated neurons through the non-linear function,so it has a great influence on the network performance.However,the selection of the activation function is a problem.Different activation functions have different advantages and disadvantages.It takes a lot of time and effort to determine the optimal activation function.In this paper,we propose a convolutional neural network based on Relu-Softplus and experiment with the handwritten digital font MNIST data setting to verify its performance,and compare it with other different activation functions.In this paper,the error rate of the image classification and the speed of convergence are analyzed,and the performance of the convolutional neural network is optimized,to solve the problem of determining the optimal activation function.There are two common learning methods in convolutional neural networks.There are supervised and unsupervised learning methods.Supervised learning is to learn the mapping function from the trained training samples,but it requires a lot of training samples,and is prone to fit.Unsupervised learning does not require training samples with labels,hoping to learn more abstract features of the hidden structure,but has a long training time,the cumbersome training process and other shortcomings.In this paper,we propose a convolutional neural network based on K-means algorithm,and experiment on the CIFAR-10 data set to verify its performance,analyzing and comparing the different network frameworks to the impact of the image classification precision.Finally,this paper applies the convolutional neural network to the road sign recognition system,and designs a road sign recognition system,which is described from the aspects of system demand analysis,outline design,detailed design and implementation.And the convolutional neural network based on K-means algorithm proposed in this paper is applied to the road sign recognition system.Finally,the training is tested on the GTRSB data set and compared with other well-known algorithms to verify the convolutional neural network based on the K-means algorithm has a certain improvement in the accuracy,reliability and timeliness of road sign classification in the application of landmark recognition system.