面向可穿戴多模生物信息传感网络的栈式自编码器优化情绪识别

发布时间：2018-01-09 00:12

本文关键词：面向可穿戴多模生物信息传感网络的栈式自编码器优化情绪识别　出处：《计算机学报》2017年08期 　论文类型：期刊论文

【摘要】：情绪识别是指采用无生命的传感器和计算机感知测量识别人类情绪状态,其主要环节包括情绪相关信号获取、特征提取以及分类识别.情绪识别可为人类情绪健康监测乃至情绪相关心理精神疾病的初筛提供科学依据.该文构建了多模可穿戴生物信息传感网络测量被测个体的多模情绪相关信号(脑电、脉搏以及血压),经由身体主站将信号传输至远程网络数据中心,并将情绪识别的结果进行网络发布,简化了测量结构,使得被测个体日常情绪监测和远程监控成为可能.由于信号测量和特征提取过程中存在不确定性,该文提出了栈式自编码器(基于深度学习理论)优化的情绪识别算法.71天时间跨度的实验结果表明,栈式自编码器预学习后的特征向量具有更高的一致性与可分性,情绪识别率较相关研究提高了约5%.
[Abstract]:Emotion recognition refers to the use of inanimate sensor and computer perception measuring and recognizing human emotional state, which includes the emotion related signal acquisition, feature extraction and classification. For emotion recognition of human emotions and emotional health monitoring related mental illness screening to provide a scientific basis. This paper constructs a multi-mode multimode signal related to the mood Wearable biological information sensor network by measuring the individual (EEG, pulse and blood pressure), the body through the master signal is transmitted to a remote network data center, and the emotion recognition results of network publishing, simplifying the measurement structure, the measured individual daily mood monitoring and remote monitoring is possible because of the uncertainty of signal. The measurement and feature extraction process, this paper from the encoder (learning theory based on stack depth) optimization of emotion recognition algorithm.71 Day time span. The experimental results show that the feature vector from the pre stack encoder after learning is consistent with the separability of higher recognition rate compared with the related studies, the mood is improved by about 5%.

【作者单位】：清华大学精密仪器系精密测试技术及仪器国家重点实验室;
【基金】：国家自然科学基金(61472216) 教育部博士点基金(20120002110067) 精密测试及仪器国家重点实验室研究项目资助~~
【分类号】：TN762;TP212
【正文快照】： 密测试及仪器国家重点实验室研究项目资助.戴逸翔,男,1990年生,博士研究生,主要研究方向为生物信息感知网络与认知信号测量解析.E-mail:daiyx12@mails.tsinghua.edu.cn.王雪(通信作者),男,1963年生,博士,教授,博士生导师,主要研究领域为精密测试传感技术和无线传感网络测量,E-

【参考文献】