基于便携式采集设备脑电信号的疲劳检测算法研究

发布时间：2018-09-01 21:02

【摘要】：疲劳是一种由长期高强度的体力劳动、脑力劳动引起的主观不适感觉,具体表现为注意力难以集中、警觉度下降、反应迟钝等症状,给正常的工作与生活带来了不便,严重时还会引起慢性疲劳症候群等疾病。因此有必要对疲劳状态进行检测并预防其危害发生。脑电信号是生物体脑部皮层神经细胞自发产生的生物电信号,与脑部活动密切相关,能够直接反映生物体的精神状态,可用作精神疲劳的评价指标。本文基于单通道便携式的脑电信号采集设备,针对疲劳检测算法进行了研究。论文的主要工作如下:1.现有大多数脑电信号相关应用研究大多基于多通道湿电极的采集设备,存在设备操作复杂、成本较高等使用局限性,本文则采用便携式干电极的采集设备,采集构建了 100人规模的数据库,包含三种精神状态的脑电信号:清醒、轻度疲劳和重度疲劳状态,该数据库可用于疲劳检测算法的训练、测试和评价;2.提出了两种实用的脑电信号预处理方法,第一种方法使用了 3Hz的高通滤波器以及50Hz陷波滤波器,能够满足疲劳检测算法的实时性和低计算复杂度要求,第二种方法在此基础上增加了集合经验模态分解过程,分解脑电信号后滤除伪迹成分,能够满足更高准确率的要求;3.针对脑电信号对应疲劳状态的特征进行了分析和实验,构建了一个融合使用时域、频域、时频域以及非线性动力学分析等方法的特征提取框架,提取45维特征,并进行了特征降维;4.根据提取的脑电信号特征,提出了一种基于深度信念网络(Deep Belief Network)的脑电信号疲劳检测算法,使用20-50-10-3结构的深度信念网络作为疲劳检测的分类器,在三种精神状态划分任务上达到了 92.55%的准确率。本实验将DBN与传统分类器、提升方法等进行比较,验证了DBN用作疲劳检测分类器的准确性和普适性。
[Abstract]:Fatigue is a kind of subjective discomfort caused by long-term, high-intensity physical labor and mental labor. It is manifested in the following symptoms: difficulty in concentrating attention, decreased alertness, slow response, and so on, which brings inconvenience to normal work and life. Serious also can cause chronic fatigue syndrome and other diseases. Therefore, it is necessary to detect the fatigue state and prevent its harm. EEG signal is a kind of bioelectric signal produced spontaneously by cortical nerve cells of the brain. It is closely related to brain activity and can directly reflect the mental state of the organism. It can be used as an evaluation index of mental fatigue. Based on single channel portable EEG signal acquisition equipment, fatigue detection algorithm is studied in this paper. The main work of the thesis is as follows: 1: 1. Most of the existing research on EEG related applications is based on multi-channel wet electrode acquisition equipment, which has some limitations, such as complex operation and high cost. In this paper, portable dry electrode acquisition equipment is used. A database of 100 people was collected and constructed, including three kinds of EEG signals of mental state: awake, mild fatigue and severe fatigue. The database can be used to train, test and evaluate fatigue detection algorithm. Two practical preprocessing methods of EEG signal are proposed. The first method uses 3Hz high-pass filter and 50Hz notch filter, which can meet the requirements of real-time and low computational complexity of fatigue detection algorithm. The second method adds the set empirical mode decomposition process to decompose the EEG signal and filter the artifacts, which can meet the requirement of higher accuracy. In this paper, the corresponding fatigue state characteristics of EEG signals are analyzed and experimented. A feature extraction framework using time-domain, frequency-domain, time-frequency and nonlinear dynamic analysis is constructed to extract 45 dimensional features. The characteristic dimension reduction is also carried out. According to the extracted EEG features, an EEG fatigue detection algorithm based on depth belief network (Deep Belief Network) is proposed, which uses 20-50-10-3 depth belief network as the classifier for fatigue detection. The accuracy rate of 92.55% has been achieved on the three mental state division tasks. By comparing DBN with traditional classifier and lifting method, the accuracy and universality of DBN as fatigue detection classifier are verified.
【学位授予单位】：北京邮电大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：R318;TN911.7

【相似文献】