The Research On Recognition Of Handwritten Digital Sequence Based On Deep Learning

Image Recognition technology plays an important role in many fields.It has widely used in machine vision,medical imaging,remote sensing images and intelligent transportation and other fields.Hand-written digits recognition is part of Image Recognition technology.When enterprises or universities research topics related to the direction of Image Recognition,they will introduce much hand-written digital recognition technologies.Hand-written digit sequences recognition is widely used in people's daily life.Therefore,our work efficiency can be better improved if the recognition accuracy and speed of the hand-written digit sequence recognition are further improved.The extraction of features in traditional digital recognition methods relies on human participation,so that the recognition algorithm has slow speed and low accuracy.While the extraction of feature in deep learning technology no longer requires manual participation,so that we can get faster speed and accuracy.This makes deep learning to become a popular research algorithms.This paper uses a deep learning method to design a system for recognizing a handwritten digital sequence and applies it to actual competitions.At first,this paper describes the research background of the topic,introduces the development status of hand-written digit recognition technology at home and abroad,and briefly introduces the machine learning technology.Then introduces the classic methods for hand-written digit sequence recognition in recent years,and focuses on the content of this study---the hand-written digital sequence recognition research based on deep learning,focuses on the deep learning algorithm.This paper gives the basic idea of deep learning and discusses the structure and training process of deep learning model based on convolutional neural network.The paper focuses on the deep learning model using Caffe as a framework and explains the core modules and basic hierarchical structure of Caffe,which includes the features of Caffe relative to other frameworks.Lenet-5 was explained as the most classic model of convolutional neural networks in this paper.And based on its improvement research,a nine-layer neural network model named Lenet-improve was designed and tested.The test proved that the improved model can improve the accuracy of digital recognition.At the end of this paper,the actual design of the Linux system,the installation and configuration of Caffe under Linux,and the detailed design process of the system's overall architecture are presented.The detailed digital image preprocessing process is given and the implementation of hand-written digital sequence recognition based on the mnist.Finally,the system is tested by the video data in the actual competition,and the result shows that the system has a high accuracy for the recognition of hand-written digit sequences.