Research And Application Of Handwritten Digit Recognition Based On Deep Learning

Handwritten Numeral Recognition(HNR)is a branch of optical character recognition,It handles handwritten Arabic digits(such as 0,1,2,....,9)by computer,allowing the computer to automatically recognize handwritten digits.According to the acquisition method,hand-written digital recognition can be divided into online recognition and offline recognition.Online recognition identifies numbers written on a handwriting device connected to a computer.The current research is mature and has been applied in various fields.Although some achievements have been made in the offline recognition study,there is still some distance from practicality as for as accuracy rate.The main research and application made in this paper are for offline handwritten numbers.The application of deep learning to the recognition of handwritten numbers is now a hot topic.Researchers at home and abroad have done a lot of research work and put forward many algorithms.However,there is sitll further improvement in recognizing and predicting the accuracy of unknown numbers.In this paper,through the study of deep learning,it is applied to the review and statistics of test paper scores.The main contents are summarized as follows:1.The research describes the background and development status of handwritten numerical recognition.Also compares and analyes the handwritten numerical recognition technique.To conclude that the handwritten numerical recogniton algorithm for deep learning is superior to other methods.2.A deep learning algorithm was studied in this paper and different depth learning models were described in detail.Meanwhile analyzing the classification of output results of the test scores to be achieved.Above work laid the foundation for the selection of the next model.3.This paper constructes a handwritten numerical recognition network model based on convolutional neural network,and introduces a paper score review system.At last,describes the necessary development tools and programing algorithoms in detail.4.The MNIST data set was used to train the constructed model,to optimize and improve the model parameters.identifying and testing the model in the self-built test set.The experimental test results show that the system design has basically reached the practical requirements of paper score recognition.