CASIA OpenIR  > 复杂系统管理与控制国家重点实验室  > 复杂系统研究
基于EEG的3D视疲劳机制与建模方法研究
岳康
Subtype博士
Thesis Advisor王丹力
2019-12
Degree Grantor中国科学院自动化研究所
Place of Conferral中国科学院自动化研究所
Degree Discipline计算机应用技术
KeywordEeg 视疲劳 卷积神经网络 Cnn
Abstract

3D显示技术能够极大的提升用户的观看体验,因而具有巨大的发展潜力和广阔的市场前景。然而,观看3D内容时出现的诸如头晕、恶心、出汗、呕吐等视疲劳症状严重影响了用户体验,并逐渐成为3D产业发展普及的瓶颈。因此,研究3D视疲劳评估方法,实现对3D视疲劳的准确、合理评估具有重要的现实意义,并逐渐成为学界研究的一个热点问题。在这些研究中,EEG以其包含丰富、全面的用户生理信息受到了广泛的关注。本文针对现有研究的不足,围绕基于EEG的3D视疲劳机制分析与建模方法开展了一系列的创新研究。论文的具体工作如下:

(1)针对当前基于EEG的视疲劳研究中缺乏相关机理研究的现状,提出使用改进的视差诱发电位技术对同3D视疲劳相关的大脑区域进行分析定位的方法。设计并实施视疲劳诱发实验,通过对信号进行时域、时频域分析,分别研究了潜伏期、振幅以及功率等EEG特征同3D视疲劳之间的关系。确定了同3D视疲劳相关的DVEP成分在大脑中的位置。通过综合分析已有文献,从解剖学角度给出受3D视疲劳影响的双目视觉信息处理过程。

(2)针对多级视疲劳分类中的类别不均衡现象,提出使用观看时长代替3D 视疲劳等级作为特征选取的依据。设计并实施基于随机点立体图的视疲劳诱发实验。通过对信号进行频域以及定位分析,研究EEG功率同观看时长之间的关系,并对关联的头皮及大脑区域进行定位。
针对EEG信号非平稳特性造成的分类模型精度较低的问题,研究基于指数滑动平均的自适应特征优化方法。研究基于EEG信号时域特征的3D视疲劳建模方法,提出基于类别加权的支持矩阵机算法用于3D视疲劳分类。

(3)针对传统分类方法中过度依赖分类特征的现象,研究基于深度学习技术的3D视疲劳评价方法。通过文献调研,综合分析基于EEG的深度学习算法给出其优缺点,确定相应的基线方法。研究基于深度学习框架的多尺度EEG时-频-空间特征抽取方法,确定抽取对应特征的网络结构并构建网络模型。通过同基线方法对比,验证提出模型的有效性,并给出相应分析。研究基于EEG的深度学习模型可视化方法,对所提出网络进行可视化验证。

Other Abstract

Three-dimensional (3D) display technology can greatly improve the user's viewing experience, thus it has great development potential and broad market prospects. However, visual fatigue symptoms such as dizziness, nausea, sweating and vomiting appear when people viewing 3D content. These undesirable symptoms provide negative user experience and gradually become the bottleneck of the development and popularization of the 3D industry. Therefore, it is of great practical significance to research the 3D visual fatigue evaluation method by which 3D visual fatigue can be evaluated accurately and reasonably. Nowadays, 3D visual fatigue has been most commonly studied using electroencephalography (EEG) since it contained rich information about the underlying neural activities and is closely associated with mental and physical activities. In this thesis, a series of innovative research has been carried out on the neural mechanism and modeling method of 3D visual fatigue based on EEG. The details are described as follows:

(1) Aiming to the lack of neural mechanism research in current studies on EEG-based visual fatigue evaluation, an improved disparity visual evoked potential (DVEP) technique was proposed to analyze and locate brain regions related to 3D visual fatigue. The relationship between latency, amplitude and power of EEG and 3D visual fatigue was studied by time-domain and time-frequency domain analysis. The location of DVEP components associated with 3D visual fatigue in the brain was identified. Based on the comprehensive analysis of existing literature, the processing processes of binocular visual information affected by 3D visual fatigue is presented from the perspective of anatomy.

(2) Aiming to the class imbalance in the classification of 3D visual fatigue, the statistical relationship between the EEG power density spectrum and viewing duration has been studied by frequency domain analysis of EEG. Moreover, scalp and cortex regions that are associated with 3D visual fatigue have also been located. Then, a feature optimization method based on the exponential moving average is proposed. Our model achieves desirable performance on our 3D visual fatigue dataset.

(3) We present a multi-scale convolutional neural network (CNN) architecture named MorletInceptionNet to detect visual fatigue using raw EEG as input, which exploits the spatial-temporal-frequency structure of EEG signals. Then, we compare our method with five state-of-the-art methods and the results demonstrate that our model achieves overally the best performance for two widely used evaluation metrics, i.e., classification accuracy and kappa value. Furthermore, we use input-perturbation network-prediction correlation maps to conduct an in-depth analysis of the reason why the proposed method outperforms other methods.

Pages134
Language中文
Document Type学位论文
Identifierhttp://ir.ia.ac.cn/handle/173211/28366
Collection复杂系统管理与控制国家重点实验室_复杂系统研究
Recommended Citation
GB/T 7714
岳康. 基于EEG的3D视疲劳机制与建模方法研究[D]. 中国科学院自动化研究所. 中国科学院自动化研究所,2019.
Files in This Item:
File Name/Size DocType Version Access License
基于EEG的3D视疲劳机制与建模方法研究(9753KB)学位论文 开放获取CC BY-NC-SAApplication Full Text
Related Services
Recommend this item
Bookmark
Usage statistics
Export to Endnote
Google Scholar
Similar articles in Google Scholar
[岳康]'s Articles
Baidu academic
Similar articles in Baidu academic
[岳康]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[岳康]'s Articles
Terms of Use
No data!
Social Bookmark/Share
All comments (0)
No comment.
 

Items in the repository are protected by copyright, with all rights reserved, unless otherwise indicated.