Comparison between CTC-based and Attention-based Methods for
发布时间:2021-04-03 02:05
化学式,作为一种直观、易于理解的知识表示模型,在化学学科教育和学术交流研究中的作用至关重要。目前,将化学式输入电子设备仍依赖于传统的点击-拖动单元组件的方式。然而,该方法既不方便,也缺乏效率。现在,随着触屏电子设备的大范围普及,手写输入对人机交互产生了很大的影响,它已经成了很多用户的首选。因此,能够直接在把化学式手写输入到电子设备中,引起了研究人员的广泛兴趣。手写化学式识别是一项有挑战的研究,由于它存在以下困难:1)是手写体的随意性,同一字符书写多样化,并且有同一字符笔迹中断、不同字符笔迹相连、出现多余的点或线条等现象存在;2)化学式长度不一样,比如单质C只有1个元素,而碱式碳酸铜(Ca2(OH)2C03)却有多达1 1个元素;3)化学式中存空间信息,如下标元素个数。成功识别手写化学式,从理论上来说,可以促进手写化学方程式的识别;另一方面,可以促进其在不同场景中的应用,如建立化学式知识库,让手写化学式搜索更加方便快捷,又如充当教学辅助。根据输入格式的不同,手写化学式识别可以分为在线和离线两个领域。在线识别中,输入数据是一系列笔画,而在离线领域中,输入数据就是含有手写化学式的图像。本文将...
【文章来源】:华中师范大学湖北省 211工程院校 教育部直属院校
【文章页数】:69 页
【学位级别】:硕士
【文章目录】:
Abstract
1 Introduction
1.1 The problem of handwritten chemical formulae recognition
1.2 Research motivations
1.3 Objectives
1.4 Contributions of the work
1.5 Thesis organisation
2 Related works
2.1 Literature review for handwritten chemical notations recognition
2.2 Artificial neural network
2.2.1 Convolutional neural network
2.2.2 VGGNet
2.2.3 Recurrent neural network
2.2.4 Long short-term memory network
2.2.5 Transfer learning
2.3 Connectionist temporal classification technique
2.4 Attention model
3 Handwritten chemical formulae recognition using CTC technique
3.1 Feature sequence extractor
3.2 Sequence dependency representation
3.3 Transcription
3.4 Objective function
4 Handwritten chemical formulae recognition using attention model
4.1 Feature sequence extractor
4.2 RNN encoder
4.3 Decoder with attention mechanism
4.4 Objective function
5 Experimentation
5.1 Data set
5.1.1 Chemical formulae selection
5.1.2 Procedure of collecting the samples
5.1.3 Data pre-process
5.2 Experimentation on handwritten chemical formulae recognition using CTC-based method
5.2.1 Experiment process
5.2.2 Experiment results
5.2.3 Analysis of prediction results
5.3 Experimentation on handwritten chemical formulae recognition using attention-based method
5.3.1 Training process
5.3.2 Experiment results
5.3.3 Visualization of attention on the dataset
5.4 Comparative Evaluation
6 Conclusions and perspectives
6.1 Conclusions
6.2 Limitations
6.3 Future work
References
Appendix A 摘要
Appendix B Acknowledgements
本文编号:3116399
【文章来源】:华中师范大学湖北省 211工程院校 教育部直属院校
【文章页数】:69 页
【学位级别】:硕士
【文章目录】:
Abstract
1 Introduction
1.1 The problem of handwritten chemical formulae recognition
1.2 Research motivations
1.3 Objectives
1.4 Contributions of the work
1.5 Thesis organisation
2 Related works
2.1 Literature review for handwritten chemical notations recognition
2.2 Artificial neural network
2.2.1 Convolutional neural network
2.2.2 VGGNet
2.2.3 Recurrent neural network
2.2.4 Long short-term memory network
2.2.5 Transfer learning
2.3 Connectionist temporal classification technique
2.4 Attention model
3 Handwritten chemical formulae recognition using CTC technique
3.1 Feature sequence extractor
3.2 Sequence dependency representation
3.3 Transcription
3.4 Objective function
4 Handwritten chemical formulae recognition using attention model
4.1 Feature sequence extractor
4.2 RNN encoder
4.3 Decoder with attention mechanism
4.4 Objective function
5 Experimentation
5.1 Data set
5.1.1 Chemical formulae selection
5.1.2 Procedure of collecting the samples
5.1.3 Data pre-process
5.2 Experimentation on handwritten chemical formulae recognition using CTC-based method
5.2.1 Experiment process
5.2.2 Experiment results
5.2.3 Analysis of prediction results
5.3 Experimentation on handwritten chemical formulae recognition using attention-based method
5.3.1 Training process
5.3.2 Experiment results
5.3.3 Visualization of attention on the dataset
5.4 Comparative Evaluation
6 Conclusions and perspectives
6.1 Conclusions
6.2 Limitations
6.3 Future work
References
Appendix A 摘要
Appendix B Acknowledgements
本文编号:3116399
本文链接:https://www.wllwen.com/kejilunwen/shengwushengchang/3116399.html
最近更新
教材专著