基于深度学习的英文字符识别方法的研究 (1).docx

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

Abstract

Withthecontinuousdevelopmentofartificialintelligence,intelligentgradingsystemshaveenteredpeoplesteachinglives.DeeplearningimageprocessingtechnologyhasmadebreakthroughprogressinCursivecharacterrecognition.Traditionalcharacterrecognitiontechnologyismainlyaimedathigh-qualityBlocklettersimages,whileinEnglishCursive,thereareirregularhandwriting,overlappingletters,wordsmearingandotherphenomena,whichcannotachievetherecognitionaccuracyofBlockletterscharacters.Inparticular,thedatabasecorrespondingtothescrawledandsmearedCursiveimagesisinsufficient,whichincreasesthedifficultyofmodeltraining.Therefore,Englishcharacterrecognitionisaverychallengingtask.

Inresponsetotheexistingproblemsmentionedaboveandthelimitationsoftraditionalhandwrittencharacterrecognitionmethods,thispaperstudiesEnglishcharacterrecognitionmethodsbasedondeeplearningtechnology,andimplementsanEnglishcharacterimagedetectionandrecognitionsystemonthisbasis.Inthispaper,adeeplearningmodelofmulti-layerConvolutionalneuralnetworkisproposedthroughthedeeplearningalgorithm.Thispaperadoptsadeeplearningmodelofmulti-layerConvolutionalneuralnetworkthroughdeeplearningalgorithm.Atpresent,theaccuracyofthismodelcanreach89%,andablationexperimentsfortheoptimizerhavebeendone.Theexperimentalresultshaveshownthatcharacterrecognitionhasahighaccuracy,butthereisstillroomforimprovementinidentifyingcharactercapitalizationandotherissues.

Keywords