皮肤电信号下学习焦虑的识别与调节技术研究

发布时间：2018-06-12 06:58

本文选题：学习焦虑 + 情感识别　；参考：《西南大学》2017年硕士论文

【摘要】：在心理学和教育学领域的研究中,学习焦虑一直受到广大学者和研究人员的重视,但是也存在很多的问题。目前,还不能够准确定量的反映出个体是否存在学习焦虑情绪,大多是通过量表和生理检测手段对学习焦虑状态进行定性的刻画,缺乏能够定量分析的可靠方法和技术手段。同时,对学习焦虑缺乏有效的实时监测与调节策略改善,这使得调节者不能准确掌握个体的适应度、接受度,这些都极大的降低了情感调节的可操作性。由于生理信号的客观真实性,基于生理信号的学习焦虑识别成为了情感计算领域的重要研究方向。本文主要是利用设计的一套实验方案来采集被试的皮肤电信号,并针对皮肤电的学习焦虑识别提出了一种改进的离散二进制粒子群算法做特征选择,并使用BP神经网络做学习焦虑的识别;在情感调节方面,参考Gross的情感调节模型,提出了人机交互环境下学习焦虑的调节模型,最后根据研究的理论成果设计和开发了基于Android的学习焦虑识别与调节助手。具体工作如下:(1)实验数据采集和特征提取。根据采集设备Shimmer3 GSR的特点以及情感诱发方案设计了一个采集学习焦虑的GSR信号的实验,实验设置两组实验场景,一组实验场景为模拟的外语课堂环境,该组主要是采集被试的学习焦虑的10min数据;另一组为观看轻松、缓和的视频的环境,该组主要是采集被试正常状态下的10min数据。接下来进行数据预处理,即根据被试在实验过程中反应,截取了20S被试在学习焦虑状态下和非学习焦虑状态下的信号,实验过程中,总共有43个被试参加,经筛选,最终有35个被试的实验数据合格,即35个学习焦虑的实验样本数据以及35个非学习焦虑的样本数据。因此本文实验中产生了70个原始样本数据。然后对原始样本进行小波去噪以作为新的样本,并提取每个样本的30个时频域统计特征。(2)特征组合优化与学习焦虑的识别。主要是建立特征优化模型和分类器模型。在特征组合优化过程中,本文采用了离散二进制粒子群优化算法(PSO),并从增强粒子多样性、提高收敛速度以及跳出局部最优等方面改进了离散二进制粒子群算法;在学习焦虑的识别过程中,采用了BP神经网络作为识别模型,并在基础上确定了特征优化的适应度目标函数。最终给出了特征组合优化结果和学习焦虑的识别结果。实验结果表明改进的粒子群算法选择的最优特征子集在BP神经网络中收敛效果要好、识别率较高。(3)建立学习焦虑的调节模型。建立了基于Gross情感调节模型的人机交互下的学习焦虑调节模型,这种调节模型能从环境控制、注意力改变、用户认知重评、用户能力与表达抑制等方面综合考虑并对学习焦虑情绪进行调节。(4)基于Android的学习焦虑识别与调节助手的设计与实现。即利用便携式GSR采集设备,实时采集用户的GSR数据到手机上,并根据前面关于学习的识别与调节理论,设计和实现一个App用于实时监控用户学习焦虑状态,如果存在学习焦虑情绪就对用户进行调节。
[Abstract]:In the field of psychology and education, learning anxiety has been paid much attention by many scholars and researchers, but there are many problems. At present, it is not able to accurately quantify whether the individual has learning anxiety or not, most of which are qualitatively depicted by the scale and physiological test. There is a lack of reliable and technical methods for quantitative analysis. At the same time, the lack of effective real-time monitoring and adjustment strategies for learning anxiety makes the regulators fail to accurately grasp the adaptability and acceptability of the individual, which greatly reduce the maneuverability of emotional adjustment. The recognition of learning anxiety has become an important research direction in the field of emotional computing. This paper mainly uses a set of experimental scheme designed to collect the skin electrical signals of the subjects, and proposes an improved discrete binary particle swarm optimization algorithm for the learning anxiety recognition of skin skin, and uses the BP neural network to do the learning focus. In the aspect of emotion regulation, referring to the emotion regulation model of Gross, the adjustment model of learning anxiety under the human-computer interaction environment is put forward. Finally, according to the theoretical results of the research, the Android based learning anxiety identification and adjustment assistant are designed and developed. The specific work is as follows: (1) experimental data acquisition and feature extraction. The characteristics of Shimmer3 GSR and the emotional induction scheme designed a GSR signal for learning anxiety. The experiment set two sets of experimental scenes, one set of experimental scenes as a simulated foreign language classroom environment, and the group was mainly the 10min data collecting the learning anxiety of the subjects; the other group was the environment for watching the relaxed and relaxed video. If we collect the 10min data in the normal state of the subjects, then the data preprocessing, that is, according to the reaction in the experiment, intercepts the signals under the learning anxiety state and the non learning anxiety state of the subjects. In the experiment, there are 43 participants in the experiment. After screening, the experimental data of the 35 subjects are qualified, that is, 35. The experiment sample data of learning anxiety and the sample data of 35 non learning anxiety. Therefore, 70 original sample data are produced in this experiment. Then, the original sample is denoised by wavelet denoising as a new sample, and the statistical characteristics of 30 time and frequency domain of each sample are extracted. (2) the identification of feature combination optimization and learning anxiety. In the process of feature combination optimization, the discrete binary particle swarm optimization (PSO) is adopted in this paper, and the discrete binary particle swarm optimization (PSO) is improved from the enhancement of particle diversity, the speed of convergence and the best jump out of the local optimality. In the process of learning anxiety, the BP God is used. On the basis of the network as the recognition model, the fitness target function is determined on the basis of the feature optimization. Finally, the results of feature combination optimization and learning anxiety are given. The experimental results show that the optimal subset selection of the improved particle swarm optimization algorithm is better in the BP neural network, and the recognition rate is higher. (3) the establishment of learning is established. The adjustment model of anxiety is established. A learning anxiety regulation model based on Gross emotion regulation model is established. This model can be considered from environmental control, attention change, user's cognitive reassessment, user ability and expression inhibition. (4) learning anxiety based on Android The design and implementation of the adjustment assistant: using the portable GSR acquisition equipment, collecting the user's GSR data on the mobile phone in real time, and designing and implementing a App to monitor the user's learning anxiety in real time according to the recognition and adjustment theory of the previous learning, and adjust the user if there is a learning anxiety.
【学位授予单位】：西南大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：G442;TN911.7;TP18

【参考文献】