The Design And Implement Of Image Digital Recognition System Based On Convolution Neural Network

With the development of science and technology,digital identification technology is currently applied to various fields,such as license plate recognition,population entry,financial industry and so on.The use of convolution neural network to identify the number of numbers can effectively improve the recognition rate of the number,but the recognition speed and efficiency need to be further optimized and improved.Based on the neural network algorithm of machine learning,this paper deeply analyzes the common requirements of image recognition and machine learning,and takes image digital recognition as the research object,including image preprocessing and processing of image recognition technology,neural network principle and convolution neural network The paper analyzes the principle and function of convolution neural network input layer,convolution layer,excitation layer,pool layer and all connection layer,focusing on hierarchical structure connection and collocation,and briefly analyzes the existing convolution neural network LeNet5 model.First of all,for the digital identification direction to select the appropriate data source,the source of data for cleaning and formatting,LeNet5 structure on the image of the size of the size of the format and gray processing,followed by rational design LeNet5 hierarchical structure,and parameters set,Finally,the results of the experiment are analyzed by horizontal comparison,and the performance of the system is evaluated,such as accuracy,running time and efficiency,and the performance of the system is analyzed.In this paper,the open source MNIST data set is used to train and study samples.The real-time handwritten digital recognition system is designed and developed.The main research work is as follows:(1)studied the use of convolution neural network for digital identification.Taking the digital recognition problem as the main object,the classification algorithm and application scene of the convolution neural network are analyzed,the machine learning environment is built,the MNIST data set is used as the test case,and the digital recognition system is constructed.(2)Based on LeNet5 convolution neural network,this paper proposes a parallel processing scheme based on GPU to reduce the training and learning time of convolution neural network.(3)The MNIST data set is used to establish the benchmark test for the digital recognition.The digital recognition of the convolution neural network algorithm is evaluated by self comparison and comparison with other algorithms.