WO2020151144A1 - 一种基于广义一致性构建脑功能网络与相关向量机的疲劳分类方法 - Google Patents

一种基于广义一致性构建脑功能网络与相关向量机的疲劳分类方法 Download PDF

Info

Publication number
WO2020151144A1
WO2020151144A1 PCT/CN2019/088444 CN2019088444W WO2020151144A1 WO 2020151144 A1 WO2020151144 A1 WO 2020151144A1 CN 2019088444 W CN2019088444 W CN 2019088444W WO 2020151144 A1 WO2020151144 A1 WO 2020151144A1
Authority
WO
WIPO (PCT)
Prior art keywords
fatigue
vector machine
signal
brain
function network
Prior art date
Application number
PCT/CN2019/088444
Other languages
English (en)
French (fr)
Inventor
王洪涛
刘旭程
吴聪
唐聪
裴子安
岳洪伟
陈鹏
Original Assignee
五邑大学
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 五邑大学 filed Critical 五邑大学
Priority to AU2019424265A priority Critical patent/AU2019424265B2/en
Publication of WO2020151144A1 publication Critical patent/WO2020151144A1/zh

Links

Images

Classifications

    • AHUMAN NECESSITIES
    • A61MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61BDIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00Measuring for diagnostic purposes; Identification of persons
    • A61B5/16Devices for psychotechnics; Testing reaction times ; Devices for evaluating the psychological state
    • A61B5/18Devices for psychotechnics; Testing reaction times ; Devices for evaluating the psychological state for vehicle drivers or machine operators
    • AHUMAN NECESSITIES
    • A61MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61BDIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00Measuring for diagnostic purposes; Identification of persons
    • A61B5/24Detecting, measuring or recording bioelectric or biomagnetic signals of the body or parts thereof
    • A61B5/316Modalities, i.e. specific diagnostic methods
    • A61B5/369Electroencephalography [EEG]

Definitions

  • the invention relates to a fatigue classification method based on generalized consistency to construct a brain function network and a correlation vector machine.
  • the physiological characteristics of the driver in a normal state are different from those in a fatigue state. Therefore, it is possible to determine whether the driver is in a fatigued driving state by detecting physiological characteristics.
  • EEG electroencephalogram
  • EEG electrooculogram
  • ECG electrocardiogram
  • EMG electromyography
  • Electro-oculogram signal EOG is the electrical potential difference signal between the cornea and the retina, which can reflect changes in the direction of the line of sight, blinking and other eye movements. According to the waveform change of EOG, it is possible to clearly analyze the change of the eye state at a certain moment, so as to determine the driver's alertness at the moment and whether it is in a fatigued driving state.
  • HR human heart rate
  • HRV heart rate variability
  • EMG Electromyography
  • EEG Electroencephalograph
  • the different rhythm waves of the EEG signal can represent different states of people, and are also closely related to fatigue.
  • the EEG signals when the driver is full of energy and when the driver is fatigued have obviously different characteristics.
  • characteristic values that can characterize the driver’s fatigue such as power spectral density ratio, Shannon entropy And so on, you can evaluate and judge different degrees of fatigue.
  • the brain electricity can directly reflect the changes in the mental state of the human brain.
  • the detection methods based on eye electricity, electrocardiogram, and myoelectricity can detect driving fatigue, it is only the brain in different mental states
  • the brain realizes information interaction through the interconnection and cluster work of different regions. The state of human consciousness and behavior is not determined by a certain region alone, but is often completed by multiple regions of the whole brain.
  • the driving fatigue detection method based on power spectrum and entropy does not involve brain regional information, and it is impossible to comprehensively and systematically study the mechanism of driving fatigue.
  • the function connection network (for example: phase lag index method, phase lock value method, etc.) cannot reflect the flow of information and the causal relationship between nodes.
  • the present invention provides a fatigue classification method based on generalized consistency to construct a brain function network and a correlation vector machine. Compared with the prior art, the present invention has higher reliability and accuracy. Direction, causality, construct an effective fatigue classification network to classify the connection characteristics of brain networks in different mental states, effectively verify the results of topology research, and improve the ability to detect driving fatigue.
  • the technical scheme of the present invention is: a fatigue classification method based on generalized consistency to construct a brain function network and a correlation vector machine, including the following steps:
  • step S1 the wireless stem electrode EEG acquisition device is used to collect the EEG signals of the subjects during simulated driving, and the duration is 90 minutes, and each subject’s EEG signals during simulated driving are collected multiple times, respectively As a training signal set and a test signal set.
  • step S1 the international 10-20 standard is used to place electrodes when collecting the EEG signals of the subject during simulated driving, with a total of 24 leads.
  • step S2 the thresholds ⁇ 1 and ⁇ 2 are set .
  • the reaction time is lower than ⁇ 1
  • the time point before ⁇ 1 is marked as awake data
  • the reaction time is between ⁇ 1 and ⁇ 2
  • the data between the time points of the two thresholds is marked as an intermediate state.
  • the reaction time is higher than ⁇ 2
  • the data after the time point of ⁇ 2 is marked as a fatigue state.
  • the frequencies of the three reconstructed sub-bands are 4-8 Hz, 8-13 Hz, and 13-30 Hz, respectively.
  • step S5 electromagnetic imaging (eLORETA) technology is used to trace the signal of each sub-band of the EEG signal, and the following formula is used to locate the cerebral cortex:
  • K i is the lead field matrix
  • H is the Hessian matrix
  • W is the weight
  • is the regularization parameter
  • (KW -1 K T + ⁇ H) + means that the value of KW -1 K T + ⁇ H is positive;
  • the time window is 10S
  • the step length is 5S.
  • step S6 the statistical coupling relationship is calculated for every two nodes in the EEG signal, and the GPDC value is calculated to obtain the information flow direction between each channel to form a factor-effect connection.
  • the calculation method of GPDC is as follows:
  • x i (t) is the data of the i-th channel
  • Ak is an N-dimensional coefficient matrix
  • ⁇ 1 (t) is Gaussian distribution
  • x(tk) is K-th order channel data
  • the unidirectional index of GPDC is defined as:
  • ⁇ ij (f) represents the influence of the signal x j on the signal x i , and the value range is 0 to 1.
  • ⁇ ij (f) is equal to 0, it means that the signal x j has no influence on the signal x i and is equal to 1.
  • the influence of x j is fully applied to x i ;
  • ⁇ i represents the diagonal of the covariance matrix, Represents the Fourier transform of the multiple linear regression model.
  • the threshold value is set to 20%, and useful connections are filtered by the threshold value. If the threshold value is greater than the threshold value, it means that the connection is useful, and it is regarded as inter-channel communication.
  • the brain function network of the subjects when they are awake and fatigued is formed, and the brain function network topology structure of the subjects in different mental states can be analyzed.
  • the correlation vector machine is used as a classification method to classify the connection characteristics of the brain network of the subject in the state of awake and fatigue to realize the detection of driving fatigue.
  • the specific classification method is as follows:
  • the probability prediction of the objective function of the input variable is:
  • ⁇ n is the weight
  • K(x, x i ) is the kernel function
  • g is a Gaussian kernel.
  • RVM Value-to-distance classification by RVM achieves a classification accuracy of more than 90%, and can effectively classify different mental states such as wakefulness and fatigue.
  • the present invention constructs a brain network through a generalized consensus algorithm, regards the brain as a multi-region collaborative network, studies its information flow direction and causal relationship between nodes, and analyzes the changes in the topology of the brain network under different mental states. It reveals the mechanism of fatigue and provides a new perspective for fatigue-related research.
  • the present invention uses a filtering algorithm to perform denoising processing on the collected EEG signals to avoid interference caused by eye electricity, myoelectricity, physical activity, equipment interference, etc., and improve the signal-to-noise ratio of the brain signals.
  • the present invention uses a signal recombination algorithm to divide the data according to frequency bands, and decomposes and reconstructs data according to frequency ranges to obtain signals of three sub-bands.
  • the present invention uses electromagnetic imaging (eLORETA) technology to trace the EEG signal, traces the scalp signal to the cerebral cortex, and then uses the generalized consensus algorithm for the traced signal to obtain the flow of information between nodes, and then construct the effect After connecting the network, based on the complex network theory, analyze the topological properties of the brain network from the awake state to the fatigue state.
  • eLORETA electromagnetic imaging
  • the present invention uses a correlation vector machine to classify connection features, which can achieve a classification accuracy of more than 90%, verify the reliability of topological structure analysis, and also provide a new method for fatigue detection.
  • FIG. 1 is a flowchart of driving fatigue detection based on generalized consistency in an embodiment of the present invention
  • Fig. 2 is an electrode placement diagram of the improved international 10-20 system adopted by the embodiment of the present invention.
  • a fatigue classification method based on generalized consistency to construct a brain function network and a correlation vector machine includes the following steps:
  • EEG signals of the subjects during simulated driving through the wireless stem electrode EEG acquisition device, the duration is 90 minutes, the EEG data of 32 subjects are collected, and the EEG data of each subject is performed twice Signal acquisition, the first time as training data, the second time as testing.
  • EEG signals use the improved international 10-20 standard to place electrodes, a total of 24 leads, and the electrode placement method is shown in Figure 2.
  • the guided car on the screen will randomly issue a braking command, record the time interval between the subject's seeing the command and the response, and count the reaction time.
  • the reaction time changes are counted, and thresholds ⁇ 1 and ⁇ 2 are set .
  • the reaction time is less than ⁇ 1
  • the time point before ⁇ 1 is marked as awake data
  • the reaction time is at ⁇ 1 and ⁇ 2 between the time data between the time point where the two thresholds labeled intermediate state
  • the reaction time is higher than ⁇ 2
  • the threshold comes from training. Due to the individual differences of the subjects, the time interval threshold setting is not uniform.
  • the time interval threshold for individual subjects needs to be obtained through training before the test.
  • the calculation method of ⁇ 1 is in the process of training experiment, from the beginning of the experiment to the first time the subject is fatigued (such as yawning) or the vehicle driving path deviates from the normal running track, the reaction time is The average value;
  • the calculation method of ⁇ 2 is the average value of the response time interval during the training experiment process, the subject is fatigued (such as yawning) or the vehicle driving path deviates from the normal running track.
  • the sampling frequency of the collected data is 250 Hz.
  • eLORETA can locate the cortex through the following formula:
  • K i is the lead field matrix
  • H is the Hessian matrix
  • W is the weight
  • is the regularization parameter
  • (KW -1 K T + ⁇ H) + means that the value of KW -1 K T + ⁇ H is positive;
  • the time window is 10S
  • the step length is 5S.
  • the method of generalized consensus algorithm is used to calculate the statistical coupling relationship between the channels.
  • the present invention adopts the method of the factor effect network, by calculating the GPDC value, the information flow direction between each channel is obtained, and a factor effect connection is formed .
  • the calculation method of GPDC is as follows:
  • x i (t) is the data of the i-th channel
  • Ak is an N-dimensional coefficient matrix
  • ⁇ 1 (t) is Gaussian distribution
  • x(tk) is K-th order channel data
  • the unidirectional index of GPDC is defined as:
  • ⁇ ij (f) represents the influence of the signal x j on the signal x i , and the value range is 0 to 1.
  • ⁇ ij (f) is equal to 0, it means that the signal x j has no influence on the signal x i and is equal to 1.
  • the influence of x j is fully applied to x i ;
  • ⁇ i represents the diagonal of the covariance matrix, Represents the Fourier transform of the multiple linear regression model.
  • the effective connection matrix contains a lot of information, and not all connections are useful.
  • the threshold is set to 20%, and useful connections are filtered. If the threshold is greater than the threshold, the connection is useful, and it is regarded as inter-channel communication, and less than the threshold. Then there is no connection.
  • the brain function network of the subject is formed when the subject is awake and fatigued, and the brain function network topology structure of the subject in different mental states can be analyzed.
  • the EEG data has high time resolution and can construct a brain function network in real time. It can monitor the changes in the brain function network of subjects from awake to fatigue in real time, which is beneficial to analyze the topological properties and information flow of the brain network.
  • a correlation vector machine is used as a classification method to classify the connection features of the brain network of the subject in the awake and fatigue state, so as to realize the detection of driving fatigue.
  • the classification method of the correlation vector machine is as follows:
  • the probability prediction of the objective function of the input variable is:
  • ⁇ n is the weight
  • K(x, x i ) is the kernel function
  • g is a Gaussian kernel.
  • RVM has achieved a classification accuracy of more than 90%, which can effectively classify different mental states such as wakefulness and fatigue, and provide a new method for the classification of driving fatigue.

Landscapes

  • Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Engineering & Computer Science (AREA)
  • Medical Informatics (AREA)
  • Surgery (AREA)
  • Biophysics (AREA)
  • Pathology (AREA)
  • Psychology (AREA)
  • Biomedical Technology (AREA)
  • Heart & Thoracic Surgery (AREA)
  • Psychiatry (AREA)
  • Molecular Biology (AREA)
  • Physics & Mathematics (AREA)
  • Animal Behavior & Ethology (AREA)
  • General Health & Medical Sciences (AREA)
  • Public Health (AREA)
  • Veterinary Medicine (AREA)
  • Child & Adolescent Psychology (AREA)
  • Developmental Disabilities (AREA)
  • Educational Technology (AREA)
  • Hospice & Palliative Care (AREA)
  • Social Psychology (AREA)
  • Measurement And Recording Of Electrical Phenomena And Electrical Characteristics Of The Living Body (AREA)

Abstract

本发明提供一种基于广义一致性构建脑功能网络与相关向量机的疲劳分类方法,本发明相对于现有技术具有较高的可靠性和准确性,通过信息流动的方向、因果关系构建有效的疲劳分类网络,以对不同精神状态下脑网络的连接特性进行分类,有效验证拓扑结构研究的结果,提高对驾驶疲劳的检测能力。本发明通过广义一致性算法的方法构建脑网络,将大脑视为一个多区域协同合作的网络,研究其信息流通方向及节点间因果关系,分析不同精神状态下大脑网络的拓扑结构变化,揭示疲劳产生的机理,为疲劳相关研究提供一种新的视角。本发明利用相关向量机对连接特征进行分类,能够实现90%以上的分类精度,验证拓扑结构分析的可靠性,还为疲劳检测提供一种新的方法。

Description

一种基于广义一致性构建脑功能网络与相关向量机的疲劳分类方法 技术领域
本发明涉及一种,尤其是一种基于广义一致性构建脑功能网络与相关向量机的疲劳分类方法。
背景技术
随着经济的快速发展,汽车成为人们生活中的主要交通工具,然而,交通安全也成为社会急需解决的问题。其中疲劳驾驶是重大交通安全事故的重要诱因之一。因此,通过研究驾驶疲劳的产生及诱发的机理,检测驾驶员的生理、心理、行为等状态,对受试者的疲劳程度做出判断,有利于提高行车安全,减少因疲劳造成的交通事故。
驾驶员在正常状态下与疲劳状态下的生理特征有所差异。因此,可以通过检测生理特征判断驾驶员是否处于疲劳驾驶状态。
目前,常用基于生理特征的检测方法主要包括脑电信号(EEG)、眼电信号(EOG)、心电信号(ECG)、肌电信号(EMG)等。其中,
眼电信号EOG(Electro-oculogram)是角膜和视网膜之间的电势差信号,能够反映出视线方向变化、眨眼等眼球运动情况。根据EOG的波形变化,可以清楚地分析出在某一时刻眼睛的状态变化,从而判断此刻的驾驶员警觉程度以及是否处于疲劳驾驶状态。
正常和疲劳状态下人的心率(HeartRate,HR)变化有着很大差异,因此,可以通过心电信号ECG(Electrocardiograph)或者脉搏信号获得心率变化,来检测驾驶员的疲劳程度。一些研究人员致力于研究心率变异性(Heart Rate Variability,HRV)与驾驶员疲劳的关系,开发基于心电图的疲劳检测系统,验证ECG或脉搏信号作为疲劳检测指标的可靠性。
肌电信号EMG(Electromyography)是肌肉活动产生的生物电信号,可以反映神经、肌肉的功能状态。Hostens等人采用诱发电位方法,发现长途驾驶的驾驶员肌电有着明显变化,随着驾驶时间的增长,驾驶员肌电信号的平均频率下降,而信号幅值却大幅度增加。
脑电信号EEG(Electroencephalograph)是最常用的检测疲劳的生理信号。EEG信号的不同节律波能够表征人的不同状态,与疲劳也有着密切关系。驾驶员精力充沛时和疲劳驾驶时的脑电信号有着明显不同的特征,通过对比分析脑电波频谱的变化规律,就可以获得可以表征驾驶员疲劳程度的特征值,如功率谱密度比值、Shannon熵等,就可以对不同程度的疲劳状态做出评价和判断。
基于生理信号的驾驶疲劳检测中,只有脑电能够直接反应人的大脑的精神状态变化,基于眼电、心电、肌电等检测方法虽然能够检测驾驶疲劳,但其仅是大脑在不同精神状态下对 身体的控制的反应,无法通过相关研究揭示疲劳在脑部的演变过程以及疲劳形成的机理,促进对大脑的研究,研究疲劳产生的根源。大脑通过不同区域的相互连接和集群工作来实现信息交互。人的意识、行为等状态也并非由某个区域单独决定,而往往是由全脑的多个区域共同协作完成的。但基于功率谱、熵的驾驶疲劳检测方法没有涉及大脑区域性的信息,无法全面、系统研究驾驶疲劳产生的机理。而在脑功能网络的构建中,功能连接网络(例如:相位滞后指数方法、相位锁定值方法等)无法反映信息的流向及节点间因果关系。在现有的疲劳相关技术中,少有对疲劳与清醒状态下的脑功能网络拓扑结构进行研究,更没有对清醒与疲劳状态下的脑网络进行分类,研究其相关方法的可靠性。
发明内容
针对现有技术的不足,本发明提供一种基于广义一致性构建脑功能网络与相关向量机的疲劳分类方法,本发明相对于现有技术具有较高的可靠性和准确性,通过信息流动的方向、因果关系构建有效的疲劳分类网络,以对不同精神状态下脑网络的连接特性进行分类,有效验证拓扑结构研究的结果,提高对驾驶疲劳的检测能力。
本发明的技术方案为:一种基于广义一致性构建脑功能网络与相关向量机的疲劳分类方法,包括以下步骤:
S1)、采集受试者模拟驾驶时的脑电信号;
S2)、在模拟驾驶过程中,随机发布刹车命令,记录受试者的反应时间;
S3)、通过独立成分分析去除采集到的原始脑电信号中的眼电信号;
S4)、采用小波包变换对脑电信号信号进行分解与重构,按照频率范围重构为三个子频带(Theta,Alpha,Beta);
S5)、采用电磁成像(eLORETA)技术对脑电信号的每个子频带的信号进行溯源,将头皮信号溯源到大脑皮层上;
S6)、采用广义一致性算法的方法计算各通道间的统计学耦合关系;
S7)、通过各通道间的耦合关系,形成因效连接网络,得到清醒与疲劳时的信息流向;
S8)、确定合适的阈值,构建脑功能网络,分析其拓扑结构变换;
S9)、通过不同的连接特性,利用相关向量机对不同精神状态进行分类,判断驾驶疲劳。
进一步的,步骤S1)中,采用无线干电极脑电采集设备采集受试者模拟驾驶时的脑电信号,时长为90min,并且每位受试者模拟驾驶时的脑电信号采集多次,分别作为训练信号集和测试信号集。
进一步的,步骤S1)中,采集受试者模拟驾驶时的脑电信号时采用国际10-20标准放置电极,共24个导联。
进一步的,步骤S2)中,设置阈值θ 1和θ 2,当反应时间低于θ 1时,θ 1所在的时间点之前标记为清醒数据,当反应时间位于θ 1和θ 2之间时,两个阈值所在时间点间的数据标记为中间状态,当反应时间高于θ 2时,θ 2所在的时间点之后的数据标记为疲劳状态。
进一步的,步骤S4)中,重构的三个子频带(Theta,Alpha,Beta)的频率分别为4-8Hz、8-13Hz、13-30Hz。
进一步的,步骤S5)中,采用电磁成像(eLORETA)技术对脑电信号的每个子频带的信号进行溯源,其如下公式实现对大脑皮层的定位:
Figure PCTCN2019088444-appb-000001
其中,
Figure PCTCN2019088444-appb-000002
为第i个素体精神活动的估计,K i是导联场矩阵,H为海森矩阵,W为权值,
Figure PCTCN2019088444-appb-000003
为第i个素体的任意测试点,α为表示正则化参数、(KW -1K T+αH) +表示KW -1K T+αH的值为正时才具有意义;
对溯源后的信号设置时间窗,时间窗为10S,步长为5S。
进一步的,步骤S6)中,对脑电信号中每两个节点计算统计学耦合关系,通过计算GPDC值,得出每个通道间信息流动方向,形成因效性连接,GPDC的计算方法如下:
对于N个通道的时间序列X(t):
X(t)=[x 1(t)...x N(t)] T
其中,x i(t)为第i个通道的数据;
假定其满足一个P阶的多元线性回归模型
Figure PCTCN2019088444-appb-000004
其中,A k是一个N维系数矩阵,E(t)=[ε 1(t)...ε N(t)] T为随机误差,ε 1(t)是高斯分布,x(t-k)为第k阶通道数据;
GPDC的单方向指数定义为:
Figure PCTCN2019088444-appb-000005
其中,π ij(f)表示信号x j对信号x i的影响,取值范围时0到1,当π ij(f)等于0时,表 信号x j对信号x i完全没有影响,等于1时,则x j产生的影响完全作用到了x i上;σ i表示协方差矩阵的对角线,
Figure PCTCN2019088444-appb-000006
表示多元线性回归模型的傅里叶变换,通过计算节点中两两之间的信息流动方向,可以形成因效性连接矩阵。
进一步的,步骤S8)中,所述的阈值设置为20%,通过阙值筛选有用连接,大于阈值则说明连接有用,视为通道间联通,小于阈值则不存在连接,经过阈值的筛选之后,形成受试者清醒、疲劳时的脑功能网络,可分析受试者不同精神状态下的脑功能网络拓扑结构。
进一步的,步骤S9)中,使用相关向量机作为分类方法对受试者清醒与疲劳状态下的脑网络的连接特征进行分类,实现驾驶疲劳的检测,具体分类方法如下:
对于输入变量x,设定分类目标为0或者1,在对y(x;w)引入逻辑sigmoid链接函数σ(y)=1/1+exp(-y),使P(t|x)符合伯努利分布,得到输入变量的目标函数的概率预测为:
Figure PCTCN2019088444-appb-000007
其中,
Figure PCTCN2019088444-appb-000008
ω n为权值;K(x,x i)为核函数。
进一步的,使用径向基作为核函数,将样本从低维空间映射到高维空间中,其公式如下:
K(x,x i)=exp(-g||x-x i|| 2);
其中,g为高斯核。
通过RVM进行分类,实现了90%以上的分类精度,可以对清醒与疲劳等不同的精神状态进行有效分类。
本发明的有益效果为:
1、本发明通过广义一致性算法的方法构建脑网络,将大脑视为一个多区域协同合作的网络,研究其信息流通方向及节点间因果关系,分析不同精神状态下大脑网络的拓扑结构变化,揭示疲劳产生的机理,为疲劳相关研究提供一种新的视角。
2、本发明利用滤波算法对采集脑电信号进行去噪处理,避免到眼电、肌电、身体活动、设备干扰等导致的干扰,提高脑信号的信噪比。
3、本发明通过信号重组算法按照频带的划分,将数据按照频率范围分解、重构得到三个子频带的信号。
4、本发明使用电磁成像(eLORETA)技术对脑电信号进行溯源,将头皮信号溯源到大脑皮层上,再对溯源后的信号使用广义一致性算法,得到节点间信息的流向,然后构建因效 联接网络,之后基于复杂网络理论,对清醒状态至疲劳状态的演变进行脑网络的拓扑属性分析。
5、本发明利用相关向量机对连接特征进行分类,能够实现90%以上的分类精度,验证拓扑结构分析的可靠性,还为疲劳检测提供一种新的方法。
附图说明
图1为本发明实施例基于广义一致性驾驶疲劳检测流程图;
图2为本发明实施例采用的改进国际10-20系统电极放置图。
具体实施方式
下面结合附图对本发明的具体实施方式作进一步说明:
如图1所示,一种基于广义一致性构建脑功能网络与相关向量机的疲劳分类方法,包括以下步骤:
S1)、通过无线干电极脑电采集设备采集受试者模拟驾驶时的脑电信号,时长为90分钟,采集32位受试者的脑电数据,对每位受试者进行两次脑电信号采集,第一次作为训练数据,第二次作为测试。在脑电信号采集时,使用改进的国际10-20标准放置电极,共24个导联,电极放置方式如图2所示。
S2)、在受试者进行模拟驾驶时,由屏幕中引导车随机发出刹车命令,记录受试者在看到命令和做出反应的时间间隔,统计反应时间。为保证受试者都进入疲劳状态,统计反应时间变化,设置阈值θ 1和θ 2,当反应时间低于θ 1时,θ 1所在的时间点之前标记为清醒数据,当反应时间位于θ 1和θ 2之间时,两个阈值所在时间点间的数据标记为中间状态,当反应时间高于θ 2时,θ 2所在的时间点之后的数据标记为疲劳状态。阈值来源于训练,由于受试者的个体差异,时间间隔阈值设置不统一。因此在测试前需要通过训练获得面向个体受试者的时间间隔阈值。其中θ 1的计算方法为在训练实验的过程中,从开始进行实验到第一次受试者表现为疲劳状态(如打呵欠)或汽车行车路径偏离正常运行轨迹的时间段内,反应时间的平均值;其θ 2计算方法是训练实验过程中,受试者外在表现为疲劳状态(如打呵欠)或汽车行车路径偏离正常运行轨迹的时间段内,反应的时间间隔的平均值。采集数据的采样频率为250Hz。
S3)、对采集的脑电信号进行预处理,独立成分分析,去除眼电信号的干扰,保留有用的脑电数据,避免到眼电、肌电、身体活动、设备干扰等导致的干扰,提高脑信号的信噪比。
S4)、采用小波包变换对脑电信号信号进行分解与重构,按照频率范围重构为三个子频带(Theta,Alpha,Beta);其中,θ波的频率为4-8Hz,α波的频率为8-13Hz,β波的频 率为13-30Hz。
S5)、重构后的24个通道的信号,采用新的节点定义方法,使用eLORETA方法,对采集到的24个通道的头皮信号进行溯源,将其溯源到大脑皮层。eLORETA可通过如下公式实现对皮层的定位:
Figure PCTCN2019088444-appb-000009
其中,
Figure PCTCN2019088444-appb-000010
为第i个素体精神活动的估计,K i是导联场矩阵,H为海森矩阵,W为权值,
Figure PCTCN2019088444-appb-000011
为第i个素体的任意测试点,α为表示正则化参数、(KW -1K T+αH) +表示KW -1K T+αH的值为正时才具有意义;
对溯源后的信号设置时间窗,时间窗为10S,步长为5S。
S6)、采用广义一致性算法的方法计算各通道间的统计学耦合关系,本发明采用因效性网络的方法,通过计算GPDC值,得出每个通道间信息流动方向,形成因效性连接。GPDC的计算方法如下:
对于N个通道的时间序列X(t):
X(t)=[x 1(t)...x N(t)] T
其中,x i(t)为第i个通道的数据;
假定其满足一个P阶的多元线性回归模型
Figure PCTCN2019088444-appb-000012
其中,A k是一个N维系数矩阵,E(t)=[ε 1(t)...ε N(t)] T为随机误差,ε 1(t)是高斯分布,x(t-k)为第k阶通道数据;
由此得到系数矩阵的傅里叶变化:
Figure PCTCN2019088444-appb-000013
GPDC的单方向指数定义为:
Figure PCTCN2019088444-appb-000014
其中,π ij(f)表示信号x j对信号x i的影响,取值范围时0到1,当π ij(f)等于0时,表 信号x j对信号x i完全没有影响,等于1时,则x j产生的影响完全作用到了x i上;σ i表示协方差矩阵的对角线,
Figure PCTCN2019088444-appb-000015
表示多元线性回归模型的傅里叶变换。
S7)、通过各通道间的耦合关系,形成因效连接网络,得到清醒与疲劳时的信息流向。
S8)、因效性连接矩阵中包含大量信息,并非所有的连接都是有用的,本实施例设置阈值为20%,筛选有用连接,大于阈值则说明连接有用,视为通道间联通,小于阈值则不存在连接。
经过阈值的筛选之后,形成受试者清醒、疲劳时的脑功能网络,可分析受试者不同精神状态下的脑功能网络拓扑结构。EEG数据时间分辨率高,可实时构建脑功能网络,能够对受试者从清醒到疲劳的脑功能网络变化进行实时监控,利于分析脑网络的拓扑属性及信息流向。
S9)、本实施例采用相关向量机作为分类方法对受试者清醒与疲劳状态下的脑网络的连接特征进行分类,实现驾驶疲劳的检测。相关向量机的分类方法如下:
对于输入变量x,设定分类目标为0或者1,在对y(x;w)引入逻辑sigmoid链接函数σ(y)=1/1+exp(-y),使P(t|x)符合伯努利分布,得到输入变量的目标函数的概率预测为:
Figure PCTCN2019088444-appb-000016
其中,
Figure PCTCN2019088444-appb-000017
ω n为权值;K(x,x i)为核函数。
进一步的,使用径向基作为核函数,将样本从低维空间映射到高维空间中,其公式如下:
K(x,x i)=exp(-g||x-x i|| 2);
其中,g为高斯核。
通过RVM进行分类,实现了90%以上的分类精度,可以对清醒与疲劳等不同的精神状态进行有效分类,为驾驶疲劳的分类提供新的方法。
上述实施例和说明书中描述的只是说明本发明的原理和最佳实施例,在不脱离本发明精神和范围的前提下,本发明还会有各种变化和改进,这些变化和改进都落入要求保护的本发明范围内。

Claims (10)

  1. 一种基于广义一致性构建脑功能网络与相关向量机的疲劳分类方法,其特征在于,包括以下步骤:
    S1)、采集受试者模拟驾驶时的脑电信号;
    S2)、在模拟驾驶过程中,随机发布刹车命令,记录受试者的反应时间;
    S3)、通过独立成分分析去除采集到的原始脑电信号中的眼电信号;
    S4)、采用小波包变换对脑电信号信号进行分解与重构,按照频率范围重构为三个子频带(Theta,Alpha,Beta);
    S5)、采用电磁成像(eLORETA)技术对脑电信号的每个子频带的信号进行溯源,将头皮信号溯源到大脑皮层上;
    S6)、采用广义一致性算法的方法计算各通道间的统计学耦合关系;
    S7)、通过各通道间的耦合关系,形成因效连接网络,得到清醒与疲劳时的信息流向;
    S8)、确定合适的阈值,构建脑功能网络,分析其拓扑结构变换;
    S9)、通过不同的连接特性,利用相关向量机对不同精神状态进行分类,判断驾驶疲劳。
  2. 根据权利要求1所述的基于广义一致性构建脑功能网络与相关向量机的疲劳分类方法,其特征在于:步骤S1)中,采用无线干电极脑电采集设备采集受试者模拟驾驶时的脑电信号,模拟时长为90min,并且每位受试者模拟驾驶时的脑电信号采集多次,分别作为训练信号集和测试信号集。
  3. 根据权利要求2所述的基于广义一致性构建脑功能网络与相关向量机的疲劳分类方法,其特征在于:步骤S1)中,采集受试者模拟驾驶时的脑电信号时采用国际10-20标准放置电极,共24个导联。
  4. 根据权利要求1所述的基于广义一致性构建脑功能网络与相关向量机的疲劳分类方法,其特征在于:步骤S2)中,设置阈值θ 1和θ 2,当反应时间低于θ 1时,θ 1所在的时间点之前标记为清醒数据,当反应时间位于θ 1和θ 2之间时,两个阈值所在时间点间的数据标记为中间状态,当反应时间高于θ 2时,θ 2所在的时间点之后的数据标记为疲劳状态。
  5. 根据权利要求1所述的基于广义一致性构建脑功能网络与相关向量机的疲劳分类方法,其特征在于:步骤S4)中,重构的三个子频带(Theta,Alpha,Beta)的频率分别为4-8Hz、8-13Hz、13-30Hz。
  6. 根据权利要求1所述的基于广义一致性构建脑功能网络与相关向量机的疲劳分类方法,其特征在于:步骤S5)中,采用电磁成像(eLORETA)技术对脑电信号的每个子频带的信号进行溯源,其通过如下公式实现对大脑皮层的定位:
    Figure PCTCN2019088444-appb-100001
    其中,
    Figure PCTCN2019088444-appb-100002
    为第i个素体精神活动的估计,K i是导联场矩阵,H为海森矩阵,W为权值,
    Figure PCTCN2019088444-appb-100003
    为第i个素体的任意测试点,α为表示正则化参数,(KW -1K T+αH) +表示KW -1K T+αH的值为正时才具有意义;
    对溯源后的信号设置时间窗,时间窗为10S,步长为5S。
  7. 根据权利要求1所述的基于广义一致性构建脑功能网络与相关向量机的疲劳分类方法,其特征在于:步骤S6)中,对脑电信号中每两个节点计算统计学耦合关系,通过计算GPDC值,得出每个通道间信息流动方向,形成因效性连接,GPDC的计算方法如下:
    对于N个通道的时间序列X(t):
    X(t)=[x 1(t)…x N(t)] T
    其中,x i(t)为第i个通道的数据;
    假定其满足一个P阶的多元线性回归模型
    Figure PCTCN2019088444-appb-100004
    其中,A k是一个N维系数矩阵,E(t)=[ε 1(t)…ε N(t)] T为随机误差,ε 1(t)是高斯分布,x(t-k)为第k阶通道数据;
    GPDC的单方向指数定义为:
    Figure PCTCN2019088444-appb-100005
    其中,π ij(f)表示信号x j对信号x i的影响,取值范围时0到1,当π ij(f)等于0时,表信号x j对信号x i完全没有影响,等于1时,则x j产生的影响完全作用到了x i上;σ i表示协方差矩阵的对角线,
    Figure PCTCN2019088444-appb-100006
    表示多元线性回归模型的傅里叶变换,通过计算节点中两两之间的信息流动方向,可以形成因效性连接矩阵。
  8. 根据权利要求1所述的基于广义一致性构建脑功能网络与相关向量机的疲劳分类方法,其特征在于:步骤S8)中,所述的阈值设置为20%,通过阈值筛选有用连接,大于阈值则连接有用,视为通道间联通,小于阈值则不存在连接,经过阈值的筛选之后,形成受试者清醒、疲劳时的脑功能网络,可分析受试者不同精神状态下的脑功能网络拓扑结构。
  9. 根据权利要求1所述的基于广义一致性构建脑功能网络与相关向量机的疲劳分类方法,其特征在于:步骤S9)中,使用相关向量机作为分类方法对受试者清醒与疲劳状态下的脑网络的连接特征进行分类,实现驾驶疲劳的检测,具体分类方法如下:
    对于输入变量x,设定分类目标为0或者1,在对y(x;w)引入逻辑sigmoid链接函数σ(y)=1/1+exp(-y),使P(t|x)符合伯努利分布,得到输入变量的目标函数的概率预测为:
    Figure PCTCN2019088444-appb-100007
    其中,
    Figure PCTCN2019088444-appb-100008
    ω n为权值;K(x,x i)为核函数。
  10. 根据权利要求9所述的基于广义一致性构建脑功能网络与相关向量机的疲劳分类方法,其特征在于:使用径向基核函数,将样本从低维空间映射到高维空间中,其公式如下:
    K(x,x i)=exp(-g||x-x i|| 2);
    其中,g为高斯核。
PCT/CN2019/088444 2019-01-24 2019-05-25 一种基于广义一致性构建脑功能网络与相关向量机的疲劳分类方法 WO2020151144A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
AU2019424265A AU2019424265B2 (en) 2019-01-24 2019-05-25 Fatigue Classiffication Method Based on Brain Function Network Constructed via Generalized Consistency and Relevant Vector Machine

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201910069956.XA CN109770924B (zh) 2019-01-24 2019-01-24 一种基于广义一致性构建脑功能网络与相关向量机的疲劳分类方法
CN201910069956.X 2019-01-24

Publications (1)

Publication Number Publication Date
WO2020151144A1 true WO2020151144A1 (zh) 2020-07-30

Family

ID=66502274

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2019/088444 WO2020151144A1 (zh) 2019-01-24 2019-05-25 一种基于广义一致性构建脑功能网络与相关向量机的疲劳分类方法

Country Status (3)

Country Link
CN (1) CN109770924B (zh)
AU (1) AU2019424265B2 (zh)
WO (1) WO2020151144A1 (zh)

Cited By (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112256122A (zh) * 2020-09-25 2021-01-22 北京师范大学 基于精神疲劳的操控工效分析方法、设备及系统
CN113100776A (zh) * 2021-04-20 2021-07-13 西北工业大学 一种融合肌电和心电信号的疲劳监测系统及方法
CN113729729A (zh) * 2021-08-13 2021-12-03 北京航空航天大学 基于图神经网络和脑网络的精神分裂症早期检测系统
CN114510966A (zh) * 2022-01-14 2022-05-17 电子科技大学 一种基于图神经网络的端到端大脑因果网络构建方法
CN114578963A (zh) * 2022-02-23 2022-06-03 华东理工大学 一种基于特征可视化和多模态融合的脑电身份识别方法
CN115192043A (zh) * 2022-07-15 2022-10-18 中山大学中山眼科中心 视疲劳预见性预测的分类模型的训练方法及训练装置
CN113842152B (zh) * 2021-09-28 2023-05-30 西交利物浦大学 脑电信号分类网络训练方法、分类方法、设备及存储介质

Families Citing this family (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109770924B (zh) * 2019-01-24 2020-06-19 五邑大学 一种基于广义一致性构建脑功能网络与相关向量机的疲劳分类方法
CN110584684B (zh) * 2019-09-11 2021-08-10 五邑大学 一种驾驶疲劳相关的eeg功能连接动态特性的分析方法
CN110664400B (zh) * 2019-09-20 2020-07-17 清华大学 基于度信息的脑电特征电位溯源方法
CN111150392A (zh) * 2020-02-12 2020-05-15 五邑大学 一种有向动态脑功能网络多类情绪识别构建方法及其装置
CN111428648B (zh) * 2020-03-26 2023-03-28 五邑大学 一种脑电信号生成网络、方法及存储介质
CN111544017A (zh) * 2020-05-25 2020-08-18 五邑大学 基于gpdc图卷积神经网络的疲劳检测方法、装置及存储介质
CN111563581B (zh) * 2020-05-27 2023-08-18 杭州电子科技大学 一种基于小波相干的脑肌功能网络构建方法
CN113456078A (zh) * 2020-11-17 2021-10-01 西安邮电大学 一种基于pdc密集图传播的疲劳驾驶识别方法
CN112656373B (zh) * 2020-12-17 2022-12-16 苏州大学 基于脑网络拓扑规律的体力疲劳检测方法及系统
CN112656372B (zh) * 2020-12-17 2022-02-15 苏州大学 基于脑网络拓扑规律的脑力疲劳检测方法及系统
CN114224359B (zh) * 2021-12-18 2023-06-30 西南交通大学 一种针对高铁调度员的脑电疲劳指标判定方法

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104473636A (zh) * 2014-12-30 2015-04-01 天津大学 一种基于偏定向相干的疲劳脑网络分析方法
CN106371610A (zh) * 2016-09-23 2017-02-01 重庆金瓯科技发展有限责任公司 一种基于脑电信号的驾驶疲劳的检测方法
CN109770924A (zh) * 2019-01-24 2019-05-21 五邑大学 一种基于广义一致性构建脑功能网络与相关向量机的疲劳分类方法

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104473636A (zh) * 2014-12-30 2015-04-01 天津大学 一种基于偏定向相干的疲劳脑网络分析方法
CN106371610A (zh) * 2016-09-23 2017-02-01 重庆金瓯科技发展有限责任公司 一种基于脑电信号的驾驶疲劳的检测方法
CN109770924A (zh) * 2019-01-24 2019-05-21 五邑大学 一种基于广义一致性构建脑功能网络与相关向量机的疲劳分类方法

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
DIMITRAKOPOULOS GEORGIOS N; KAKKOS IOANNIS; DAI ZHONGXIANG; WANG HONGTAO; SGARBAS KYRIAKOS; THAKOR NITISH; BEZERIANOS ANASTASIOS; : "Functional Connectivity Analysis of Mental Fatigue Reveals Different Network Topological Alterations Between Driving and Vigilance Tasks", IEEE TRANSACTIONS ON NEURAL SYSTEMS AND REHABILITATION ENGINEERING, vol. 26, no. 4, 11 January 2018 (2018-01-11), pages 740 - 749, XP011680554, ISSN: 1534-4320, DOI: :10.1109/TNSRE.2018.2791936 *

Cited By (12)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112256122A (zh) * 2020-09-25 2021-01-22 北京师范大学 基于精神疲劳的操控工效分析方法、设备及系统
CN112256122B (zh) * 2020-09-25 2022-08-19 北京师范大学 基于精神疲劳的操控工效分析方法、设备及系统
CN113100776A (zh) * 2021-04-20 2021-07-13 西北工业大学 一种融合肌电和心电信号的疲劳监测系统及方法
CN113100776B (zh) * 2021-04-20 2023-08-29 西北工业大学 一种融合肌电和心电信号的疲劳监测系统及方法
CN113729729A (zh) * 2021-08-13 2021-12-03 北京航空航天大学 基于图神经网络和脑网络的精神分裂症早期检测系统
CN113842152B (zh) * 2021-09-28 2023-05-30 西交利物浦大学 脑电信号分类网络训练方法、分类方法、设备及存储介质
CN114510966A (zh) * 2022-01-14 2022-05-17 电子科技大学 一种基于图神经网络的端到端大脑因果网络构建方法
CN114510966B (zh) * 2022-01-14 2023-04-28 电子科技大学 一种基于图神经网络的端到端大脑因果网络构建方法
CN114578963A (zh) * 2022-02-23 2022-06-03 华东理工大学 一种基于特征可视化和多模态融合的脑电身份识别方法
CN114578963B (zh) * 2022-02-23 2024-04-05 华东理工大学 一种基于特征可视化和多模态融合的脑电身份识别方法
CN115192043A (zh) * 2022-07-15 2022-10-18 中山大学中山眼科中心 视疲劳预见性预测的分类模型的训练方法及训练装置
CN115192043B (zh) * 2022-07-15 2023-03-31 中山大学中山眼科中心 视疲劳预见性预测的分类模型的训练方法及训练装置

Also Published As

Publication number Publication date
CN109770924B (zh) 2020-06-19
AU2019424265A1 (en) 2021-07-08
AU2019424265B2 (en) 2022-02-03
CN109770924A (zh) 2019-05-21

Similar Documents

Publication Publication Date Title
WO2020151144A1 (zh) 一种基于广义一致性构建脑功能网络与相关向量机的疲劳分类方法
Kiymik et al. Automatic recognition of alertness level by using wavelet transform and artificial neural network
US8838227B2 (en) Portable automatic brain state assessment apparatus
Kumari et al. Seizure detection in EEG using time frequency analysis and SVM
Gruszczyńska et al. Epilepsy identification based on EEG signal using RQA method
US20180279938A1 (en) Method of diagnosing dementia and apparatus for performing the same
CN112190261A (zh) 一种基于静息态脑网络的孤独症脑电信号的分类装置
Wang et al. EEG characteristic analysis of coach bus drivers based on brain connectivity as revealed via a graph theoretical network
CN112426162A (zh) 一种基于脑电信号节律熵的疲劳检测方法
Wei et al. Spindle-AI: sleep spindle number and duration estimation in infant EEG
Zou et al. The functional brain network based on the combination of shortest path tree and its application in fatigue driving state recognition and analysis of the neural mechanism of fatigue driving
Bairagi et al. Epileptic seizure identification in EEG signals using DWT, ANN and sequential window algorithm
Kang et al. Accurate assessment of low-function autistic children based on EEG feature fusion
CN116211308A (zh) 一种高强度运动下机体疲劳评估方法
Liu et al. Epilepsy EEG classification method based on supervised locality preserving canonical correlation analysis
Liu et al. EEG based Mental Workload Assessment by Power Spectral Density Feature
Aghaeeaval et al. Prediction of patient survival following postanoxic coma using EEG data and clinical features
Kalbhor et al. The review of detection and classification of epilectic seizures using wavelet transform
Gao K-Means Algorithm to Identify the Elderly Psychological Stress Analysis Algorithm
Yaakop et al. Detection of drowsiness using EEG probes sensory logic signals activeness topology
Snyder Time series modeling of clinical electroencephalogram data-an information theory approach
Zhang et al. Classification of Vigilance Based on EEG
Han et al. Epileptic seizure prediction using effective brain network and svm on eeg
Olaniyan et al. Development of a microcontroller EEG-based system for diagnosis of autism spectrum disorder in developing countries
Ashtiani et al. Spectral characteristics assessment in recognition of drivers' drowsiness using statistical tests

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 19911351

Country of ref document: EP

Kind code of ref document: A1

ENP Entry into the national phase

Ref document number: 2019424265

Country of ref document: AU

Date of ref document: 20190525

Kind code of ref document: A

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 19911351

Country of ref document: EP

Kind code of ref document: A1