基于深度学习的英文字符识别方法的研究

摘要随着人工智能的不断发展，智能批阅系统走进了人们的教学生活。深度学习图像处理技术在手写体字符识别中取得了突破性的进展，传统的文字识别技术主要面向高质量的印刷体图像，而英文手写体中存在字迹不工整、字母之间重叠、单词涂抹等现象，无法达到印刷体的识别精度。特别是书写潦草、涂抹的手写体图像对应的数据库构建不足，增加了模型训练的难度，因此英文字符识别是一项极具挑战性的任务。针对上述现存的问题和传统手写文字识别方法的局限性，本文研究了基于深度学习技术的英文字符识别方法，并在此基础上实现了英文字符图像检测识别系统。本文通过深度学习算法采用一种多层卷积神经网络的深度学习模型，目前该模型的准确率可达89%，并且做了针对优化器的消融实验。实验结果证明，字符识别有较高的准确率，但是在识别字符大小写等问题上仍有提高空间。关键词：深度学习；英文字符识别；卷积循环神经网络AbstractWith the continuous development of artificial intelligence,intelligent grading systems haveentered people's teaching lives.Deep learning image processing technology has madebreakthrough progress in Cursive character recognition.Traditional character recognitiontechnology is mainly aimed at high-quality Block letters images,while in English Cursive,thereare irregular handwriting,overlapping letters,word smearing and other phenomena,whichcannot achieve the recognition accuracy of Block letters characters.In particular,the databasecorresponding to the scrawled and smeared Cursive images is insufficient,which increases thedifficulty of model training.Therefore,English character recognition is a very challenging task.In response to the existing problems mentioned above and the limitations of traditionalhandwritten character recognition methods,this paper studies English character recognitionmethods based on deep learning technology,and implements an English character imagedetection and recognition system on this basis.In this paper,a deep learning model ofmulti-layer Convolutional neural network is proposed through the deep learning algorithm.Thispaper adopts a deep learning model of multi-layer Convolutional neural network through deeplearning algorithm.At present,the accuracy of this model can reach 89%,and ablationexperiments for the optimizer have been done.The experimental results have shown thatcharacter recognition has a high accuracy,but there is still room for improvement in identifyingcharacter capitalization and other issues.Key words:Deep learning;English handwriting recognition Convolutional recurrentneural network目录摘要…IAbstract.....Ⅱ第1章绪论…1.1研究背景以及意义…12国内外研究现状以及分析….21.2.1英文字符识别国内外状况.21.2.2国内外发展现状31.3研究现状和存在问题…。41.4论文的章节安排….4第2章相关理论介绍....62.1深度学习基础。62.1.1卷积神经网络.62.1.2循环神经网络.92.1.3注意力机制…112.2数据集介绍.….142.3本章小结…15第3章基于深度神经网络英文字符识别方法163.1CNCR模型原理163.2模型优化器.173.2.1模型优化器的作用>3.2.2Adam优化器公3.3本章小结.18第4章实验结论…194.1实验结果与分析194.2优化器不同的实验结果.204.3对比试验…264.4本章小结27第5章总结与展望28参考文献…30致谢.3