CN106333680A - Sleep state detection method and system based on fusion of multiple classifiers - Google Patents

Sleep state detection method and system based on fusion of multiple classifiers Download PDF

Info

Publication number
CN106333680A
CN106333680A CN201610843526.5A CN201610843526A CN106333680A CN 106333680 A CN106333680 A CN 106333680A CN 201610843526 A CN201610843526 A CN 201610843526A CN 106333680 A CN106333680 A CN 106333680A
Authority
CN
China
Prior art keywords
sleep state
output
classifier
band
user
Prior art date
Application number
CN201610843526.5A
Other languages
Chinese (zh)
Inventor
赵巍
胡静
韩志
Original Assignee
广州视源电子科技股份有限公司
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 广州视源电子科技股份有限公司 filed Critical 广州视源电子科技股份有限公司
Priority to CN201610843526.5A priority Critical patent/CN106333680A/en
Publication of CN106333680A publication Critical patent/CN106333680A/en

Links

Classifications

    • AHUMAN NECESSITIES
    • A61MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61BDIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00Detecting, measuring or recording for diagnostic purposes; Identification of persons
    • A61B5/04Measuring bioelectric signals of the body or parts thereof
    • A61B5/0476Electroencephalography
    • AHUMAN NECESSITIES
    • A61MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61BDIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00Detecting, measuring or recording for diagnostic purposes; Identification of persons
    • A61B5/48Other medical applications
    • A61B5/4806Sleep evaluation
    • A61B5/4809Sleep detection, i.e. determining whether a subject is asleep or not
    • AHUMAN NECESSITIES
    • A61MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61BDIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00Detecting, measuring or recording for diagnostic purposes; Identification of persons
    • A61B5/48Other medical applications
    • A61B5/4806Sleep evaluation
    • A61B5/4812Detecting sleep stages or cycles
    • AHUMAN NECESSITIES
    • A61MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61BDIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00Detecting, measuring or recording for diagnostic purposes; Identification of persons
    • A61B5/48Other medical applications
    • A61B5/4806Sleep evaluation
    • A61B5/4815Sleep quality
    • AHUMAN NECESSITIES
    • A61MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61BDIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00Detecting, measuring or recording for diagnostic purposes; Identification of persons
    • A61B5/72Signal processing specially adapted for physiological signals or for diagnostic purposes
    • AHUMAN NECESSITIES
    • A61MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61BDIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00Detecting, measuring or recording for diagnostic purposes; Identification of persons
    • A61B5/72Signal processing specially adapted for physiological signals or for diagnostic purposes
    • A61B5/7203Signal processing specially adapted for physiological signals or for diagnostic purposes for noise prevention, reduction or removal
    • AHUMAN NECESSITIES
    • A61MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61BDIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00Detecting, measuring or recording for diagnostic purposes; Identification of persons
    • A61B5/72Signal processing specially adapted for physiological signals or for diagnostic purposes
    • A61B5/7235Details of waveform analysis
    • A61B5/725Details of waveform analysis using specific filters therefor, e.g. Kalman or adaptive filters
    • AHUMAN NECESSITIES
    • A61MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61BDIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00Detecting, measuring or recording for diagnostic purposes; Identification of persons
    • A61B5/72Signal processing specially adapted for physiological signals or for diagnostic purposes
    • A61B5/7235Details of waveform analysis
    • A61B5/7264Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems
    • AHUMAN NECESSITIES
    • A61MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61BDIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00Detecting, measuring or recording for diagnostic purposes; Identification of persons
    • A61B5/72Signal processing specially adapted for physiological signals or for diagnostic purposes
    • A61B5/7235Details of waveform analysis
    • A61B5/7264Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems
    • A61B5/7267Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems involving training the classification device

Abstract

The invention relates to a sleep state detection method and system based on fusion of multiple classifiers. The method comprises the following steps: acquiring an electroencephalogram of a user generated during a sleep process, and extracting corresponding characteristic data from the electroencephalogram according to a recognition task recognized by the sleep state; respectively inputting the characteristic data into a preset classifier and detectors 1 to N to perform the detection; if an output result of only one detector is true, detecting the sleep state of the user by using the output result of the detector, and identifying the type of the characteristic data; if the output results of a plurality of detectors are true, or the output results of all detectors are false, detecting the sleep state of the user by using the output result of the preset classifier; and training the preset classifier by using the identified characteristic data to obtain a new classifier, substituting the preset classifier with the new classifier, and detecting the sleep state of the user. By adopting the sleep state detection method and system, the accuracy of the classifier can be increased, and the detection accuracy of the sleep state can be improved.

Description

基于多分类器融合的睡眠状态检测方法和系统 Based sleep state detection method and system integration of multiple classifiers

技术领域 FIELD

[0001] 本发明涉及辅助睡眠技术领域,特别是涉及一种基于多分类器融合的睡眠状态检测方法和系统。 [0001] The technical field relates to a sleep aid, more particularly to a method and system for detecting a sleep state based on multi Fusion with the present invention.

背景技术 Background technique

[0002] 在睡眠中,人体进行了自我放松及恢复的过程,因此良好的睡眠是保持身体健康的一项基本条件;但是由于工作压力大、生活作息不规律等原因,导致了部分人群的睡眠质量欠佳,表现为失眠、半夜惊醒等。 [0002] In sleep, the body was self-relaxation and recovery process, so good night's sleep is a basic condition for maintaining good health; but due to work pressure, irregular lifestyle and other reasons, led to the sleeping part of the population poor quality, manifested as insomnia, middle of the night woke up and so on.

[0003] 目前市面上已经有一些设备来帮助人们入睡,提高睡眠质量。 [0003] There are already some devices to help people fall asleep and improve sleep quality. 例如在某一特定睡眠状态下通过声音、光信号等人工干预,避免在熟睡状态下叫醒用户等。 For example in a particular sleep state by sound, light signals manual intervention, to avoid wake the user in the sleeping status. 对于辅助睡眠的设备而言,为了真正达到提高用户睡眠质量的目的,正确的检测用户的睡眠状态是非常重要的。 For sleep aid equipment, in order to really achieve the purpose of improving the user's quality of sleep, proper detection of the user's sleep it is very important.

[0004] 目前临床上主要采用多导睡眠图识别睡眠状态,主要是利用脑电信号来对睡眠进行分析,通过训练睡眠状态模型来识别被测者的睡眠状态,例如判断用户处于睡眠的哪个阶段,但由于脑电信号的个人特异性很强,并且强度很弱,在信号采集时极易被外界信号所干扰。 [0004] The currently used mainly to identify polysomnographic sleep clinically mainly used to analyze the EEG sleep, sleep state to identify the subject's sleep state by training the model, for example, the user determines which stage in sleep but due to the strong individual-specific EEG, and is very weak, can easily be disturbed by external signal when the signal acquisition. 因此一般预先训练出来的分类器对很多用户的检测存在误差,准确性难以得到保证。 It is generally pre-trained classifier for many users there is an error detection accuracy is difficult to guarantee.

发明内容 SUMMARY

[0005] 基于此,有必要针对上述问题,提供一种基于多分类器融合的睡眠状态检测方法和系统,有效地提高预设分类器识别的准确性。 [0005] Based on this, it is necessary to solve these problems, there is provided a sleep state based on detection methods and systems fusion of multiple classifiers effectively improve the accuracy of identification of a predetermined classification.

[0006] -种基于多分类器融合的睡眠状态检测方法,包括: [0006] - sleep detection method based species multiple classifiers fusion, comprising:

[0007] 采集用户在睡眠过程中产生的脑电信号,根据睡眠状态识别的识别任务从所述脑电信号中提取相应的特征数据; [0007] EEG acquisition generated by the user during sleep, the recognition task identification feature data extracted from the respective EEG The sleep state;

[0008] 将所述特征数据分别输入预设分类器、检测器1~N进行检测;其中,所述预设分类器检测用户的1~N种睡眠状态,检测器1~N分别对应检测用户的一种特定睡眠状态,N多2; [0008] The predetermined classification characteristic data are input, a detector for detecting 1 ~ N; wherein said predetermined sleep state 1 ~ N kinds classifier detects a user, the detector 1 ~ N respectively detect user one particular sleep state, N number 2;

[0009] 若只有一个检测器的输出结果为真,则以该检测器的输出结果检测用户的睡眠状态,并对该特征数据进行类型标注;若有多个检测器的输出结果为真,或者全部检测器的输出结果均为假,则以预设分类器的输出结果检测用户的睡眠状态; [0009] If only one detector output is true, places the output of the detector detecting the user's sleep state, and the type of the annotation feature data; if a plurality of detector output is true, or the output of all detectors are bogus, places the user's default output detection classifier sleep;

[0010] 利用标注的特征数据对所述预设分类器进行训练得到新的分类器,并利用该新的分类器取代所述预设分类器,检测用户的睡眠状态。 [0010] labeling using the obtained feature data preset new classifier is trained classifier, and a classifier utilizing the novel substituted the preset classifier, detecting a user's sleep state.

[0011] -种基于多分类器融合的睡眠状态检测系统,包括: [0011] - Multi Species Fusion with sleep detection system, comprising:

[0012] 特征数据提取模块,用于采集用户在睡眠过程中产生的脑电信号,根据睡眠状态识别的识别任务从所述脑电信号中提取相应的特征数据; [0012] The feature data extraction means for collecting electrical signals generated by the user in the brain during sleep, the recognition task identification feature data extracted from the respective EEG The sleep state;

[0013] 分类器融合检测模块,用于将所述特征数据分别输入预设分类器、检测器1~N进行检测;其中,所述预设分类器检测用户的1~N种睡眠状态,检测器1~N分别对应检测用户的一种特定睡眠状态,N彡2; [0013] classifier fusion detection module configured to preset the input feature data are classified, detector detects 1 ~ N; wherein said predetermined sleep state 1 ~ N kinds classifier detects a user, the detection is 1 ~ N correspond to a particular user's sleep state detection, N San 2;

[0014] 结果判断和数据标注模块,用于若只有一个检测器的输出结果为真,则以该检测器的输出结果检测用户的睡眠状态,并对该特征数据进行类型标注;若有多个检测器的输出结果为真,或者全部检测器的输出结果均为假,则以预设分类器的输出结果检测用户的睡眠状态; [0014] Analyzing the results and data annotation module, if only for a detector output is true, places the output of the detector detecting the user's sleep state, and the type of the annotation feature data; if a plurality of the output of the detector is true, or the output of the detector are all false, the user places a predetermined detection output of classifier sleep state;

[0015] 分类器训练和更新模块,用于利用标注的特征数据对所述预设分类器进行训练得到新的分类器,并利用该新的分类器取代所述预设分类器,检测用户的睡眠状态。 [0015] classifier training and updating module, for utilizing feature data of the predetermined label classifier is trained classifier to obtain a new, and utilizing the new classifier classifier instead of the preset, the user is detected sleep state.

[0016] 上述基于多分类器融合的睡眠状态检测方法和系统,基于脑电信号的特征数据, 在预设分类器基础上,进一步设置了针对多个睡眠状态的检测器1~N,融合分类器检测结果和多个检测器输出结果对特征数据进行类型标注,然后通过标注类型的特征数据输入预设分类器训练出新的分类器,取代原预设分类器,检测用户的睡眠状态。 [0016] The sleep state based on detection methods and systems fusion of multiple classifiers, the EEG data based on the feature, the preset classification basis, further provided ~ N, fusion classified sleep states for a plurality of detector 1 and a plurality of detection results of detector output data type annotation feature, then a new classifier training input feature data by a predetermined classification label type, replacing the default classifier, detecting a user's sleep state. 该方案可以在使用过程中训练出更加接近于用户个人特异性的分类器,可以提高分类器的准确率,增强睡眠状态检测的准确性。 The program can be trained in the course of a closer personal user-specific classifier, can improve the accuracy of classification, and enhance the accuracy of sleep detection.

附图说明 BRIEF DESCRIPTION

[0017] 图1为一个实施例的基于多分类器融合的睡眠状态检测方法的流程图; [0017] FIG. 1 is a flow chart showing a method for detecting a sleep state of fusion of multiple classifiers according to one embodiment;

[0018] 图2为滤波处理前后的脑电信号示意图; [0018] FIG. 2 is a schematic view of the EEG signal before and after filtering;

[0019] 图3是多分类器融合检测器检测睡眠状态的示意图; [0019] FIG. 3 is a schematic view of a multi-classifier fusion sleep state detector;

[0020] 图4为一个实施例的基于多分类器融合的睡眠状态检测系统结构示意图。 [0020] FIG. 4 is a schematic structural diagram of a sleep state based on the embodiment of the multi Fusion with detection system.

具体实施方式 Detailed ways

[0021] 下面结合附图阐述本发明的基于多分类器融合的睡眠状态检测方法和系统的实施例。 [0021] annexed drawings set forth below based on the sleep state detecting method and system integration of multiple classifiers embodiment of the present invention in combination.

[0022] 参考图1所示,图1为一个实施例的基于多分类器融合的睡眠状态检测方法的流程图,包括: , The sleep state based on a method for detecting a fusion of multiple classifiers flowchart [0022] Referring to FIG 1 showing an embodiment, comprising:

[0023] 步骤S101,采集用户在睡眠过程中产生的脑电信号,根据睡眠状态识别的识别任务从所述脑电信号中提取相应的特征数据; [0023] In step S101, the user generated acquisition EEG during sleep, the recognition task identification feature data extracted from the respective EEG The sleep state;

[0024] 在本步骤中,如在对用户进行辅助睡眠时,通过用户佩戴相关传感设备,检测用户的脑电信号,在采集脑电信号时,可以以30s为一帧进行采集。 [0024] In this step, when the user such as a sleep aid, worn by a user associated sensing device detecting the user's EEG, in collecting EEG signals may be collected as a frame 30s.

[0025] 根据需要进行睡眠状态识别的任务,确定特征数据类型,从脑电信号中提取与之相应的特征数据;例如,要识别1~N种睡眠状态,提取用于进行这N种状态识别的特征数据。 [0025] The task of identifying the sleep state if necessary, to determine the type of characteristic data, extracted feature data corresponding thereto from the EEG signal; e.g., 1 ~ N species to identify the sleep state, the extraction of the N states for identification It features data.

[0026] 在一个实施例中,提取特征数据前,还可以对所采集的脑电信号进行滤波处理,滤除高频噪声和工频干扰。 Before [0026] In one embodiment, feature extraction data, may also be filtered EEG signal acquired, filtering out high frequency noise and frequency interference. 例如,脑电信号的有用信息多集中在〇-100Hz的范围内,在采集过程中会掺入频率在该范围外的噪声,因此,可以通过滤波手段将其滤除。 For example, useful information in the EEG is more concentrated in the range of billion-100Hz, the acquisition process will be incorporated in the noise outside the frequency range, therefore, it can be filtered off by the filtering means. 可以同带通滤波器滤除高频噪声,并设计一个陷波器(50/60HZ)来滤除工频干扰。 It may filter out high frequency noise with a bandpass filter, and a notch filter design (50 / 60HZ) to filter frequency interference.

[0027] 参考图2所示,图2为滤波处理前后的脑电信号示意图,上图为原始信号,下图为经过滤波处理之后的信号,可以发现大部分的高频噪声已被滤除。 [0027] Referring to FIG. 2, FIG. 2 is a schematic view of the EEG signal before and after filtering, the picture shows the original signal, the signal after the after picture shows the filtering process can be found in most of the high frequency noise has been filtered.

[0028] 对于提取特征数据的方案,本发明提供若干实施例,具体过程包括如下: [0028] For the extracted feature data, the present invention provides several embodiments, specific process includes the following:

[0029] (1)提取脑电信号的基线,计算所述基线的变化幅度;其中,所述变化幅度为基线最大值减去最小值; [0029] (1) extracting baseline EEG, calculating the variation width of the baseline; wherein said variation amplitude of the maximum minus the minimum baseline;

[0030] (2)在去掉基线后,对所述脑电信号进行小波分解,获得小波系数,并根据小波系数计算小波系数的特征参数;其中,所述特征参数包括小波系数的均值、方差、峭度系数和/ 或斜度系数; [0030] (2) after removal of the baseline, the EEG wavelet decomposition to obtain wavelet coefficients, wavelet coefficients and characteristic parameters calculated on the basis of wavelet coefficients; wherein said characteristic parameters comprise wavelet coefficients mean, variance, kurtosis coefficients and / or slope coefficient;

[0031] 为了更好地分解出所述各种频率波形,小波分解的层数与脑电信号的采样频率满足如下关系:f = 2N+2,其中,f为脑电信号的采样频率,N为小波分解的层数;例如,当信号的降采样率为128Hz时,可以选择4层分解,当信号的采样率为256Hz时,则可以进行5层分解。 [0031] In order to decompose the waveforms of various frequencies, wavelet decomposition layers and EEG sampling frequency satisfies the following relation: f = 2N + 2, where, f is the sampling frequency EEG, N wavelet decomposition layers; for example, when the downsampling rate is 128Hz signal, may be selected decomposition layer 4, when the signal sampling rate 256Hz, the layer 5 may be decomposed.

[0032] ⑶在去掉基线后,计算脑电信号的LZ复杂度和样本熵; [0032] ⑶ after removing the baseline EEG LZ calculation complexity and sample entropy;

[0033] 将所述基线的变化幅度、小波系数的特征参数、LZ复杂度和样本熵设为所述特征数据; [0033] The characteristic parameters of the baseline amplitude wavelet coefficients, LZ sample entropy and complexity to the characteristic data;

[0034] 由上述实施例的方案,作为信号特征的数据包括了基线的变化幅度、小波系数的特征参数、LZ复杂度和样本熵等。 [0034] Examples of the above embodiments, as the signal characteristic data include characteristic parameters amplitude, baseline wavelet coefficients, LZ sample complexity and entropy.

[0035] 进一步地,还可以利用脑电信号的多个波段的波形来进行识别,在小波重构中提取所述脑电信号S波频段、Θ波频段、α波频段和β波频段的信号;根据频率的不同,脑电信号是可以分为4种节律脑电波:δ波(1-3Ηζ),θ波(4-7Ηζ),α波(8-12Ηζ),β波(14-30ΗΖ),在此, 可以提取出这四种频段的信号后,利用这些信号来计算相关特征,具体方案可以如下: [0035] Further, it is also the waveform of a plurality of bands of EEG signals to be identified, extracts the S-wave EEG frequency band, the frequency band [Theta] wave, α wave signal, and β wave frequency band in wavelet reconstruction ; depending on the frequency, EEG rhythms can be classified into four brainwave: δ wave (1-3Ηζ), θ wave (4-7Ηζ), α wave (8-12Ηζ), β wave (14-30ΗΖ) after this, the band can be extracted four signals, using these signals to calculate correlation characteristics, specific programs may be as follows:

[0036] (4)分别计算脑电信号中δ波频段,Θ波频段,α波频段,β波频段的能量在总能量中的比例;将该比例也作为特征数据输入分类器进行识别;计算方法可以包括如下公式: [0036] (4) δ EEG wave band, the wave band [Theta], [alpha] frequency wave, β wave energy band proportion of the total energy were calculated; the proportion of the input data as the feature classifier recognition; calculated The method may comprise the following formula:

Figure CN106333680AD00071

[0041] 其中pt〇tai= Σ (ys) 2+Σ (ye) 2+Σ (ya) 2+Σ (yp) 2,^,7〇,7(1和7{!分别表示重构后的5频段、Θ频段、α频段和β频段的信号,^,",^和" !分别代表3频段、0频段、(1频段和3频段的信号的能量在总能量的比例。 [0041] wherein pt〇tai = Σ (ys) 2 + Σ (ye) 2 + Σ (ya) 2 + Σ (yp) 2, ^, 7〇, 7 (1 and 7 {! Represent reconstructed 5-band, the frequency band [Theta], [alpha] frequency band signal, and β, ^ ", and ^!" representing three frequency bands, bands 0, the ratio of (1 3 and the band energy of the signal in the frequency band of the total energy.

[0042] ⑶分别计算在一帧的时间内,脑电信号中δ波频段,Θ波频段,α波频段,β波频段能量最大的时间长度;将该时间也作为特征数据输入分类器进行识别,计算方法可以包括如下公式: [0042] ⑶ were calculated within the time frame, frequency, frequency band [Theta], [alpha] frequency wave, β wave energy band maximum length of time δ wave EEG wave; time as the characteristic identification data input classifier , calculation method may comprise the following formula:

Figure CN106333680AD00072

[0047] 式中,Cs,ce,Ca和Ce表示δ频段、Θ频段、α频段和β频段的信号在当前帧内所占能量比例最大的时间长度,分别表示第i秒内δ频段、Θ频段、α频段和β频段的信号的能量在总能量的比例。 [0047] In the formula, Cs, ce, Ca, and Ce represent δ band, the frequency band [Theta], [alpha] and β frequency band signal frame time length of the maximum percentage of the energy in the current ratio, respectively, within the frequency band of the i δ seconds, [Theta] the proportion of the energy band, the frequency band signals α and β of the total energy in the band.

[0048] 步骤S102,将所述特征数据分别输入预设分类器、检测器1~N进行检测;其中,所述预设分类器检测用户的1~N种睡眠状态,检测器1~N分别对应检测用户的一种特定睡眠状态,N彡2; [0048] step S102, the feature data are input preset classifier, 1 ~ N detectors for detecting; wherein said predetermined sleep state 1 ~ N kinds classifier detects a user, the detector 1 ~ N, respectively, detecting a user corresponding to a particular sleep state, N San 2;

[0049] 对于上述预设分类器,可以是采用RBF核的SVM (Support Vector Machin,支持向量机)分类器,也可以采用神经网络、决策树的分类器。 [0049] For the above-described preset classifier, may be employed RBF kernel SVM (Support Vector Machin, SVM) classifier can be employed neural network, decision tree classifier. 该分类器是通过其他样本数据训练得到。 The classifier was trained by other sample data.

[0050] 训练过程可以如下: [0050] training process can be as follows:

[0051] (1)获取所述用户的特征数据,从两种类型的特征数据中分别随机抽取相同数量的样本作为训练数据,其余作为测试数据; [0051] (1) acquiring the characteristic data of the user, respectively the same number of samples randomly selected from the two types of characteristic data as training data, test data as the remainder;

[0052] (2)将所述训练数据输入支持向量机分类器或神经网络进行自学习,利用grid-test 算法寻找识别率最高的参数,将该参数设为最优参数; [0052] (2) the data input training SVM classifier or a neural network self-learning algorithm using a grid-test to find the highest recognition rate of the parameter, the parameter is set to the optimum parameters;

[0053] 对于支持向量机分类器,训练过程中采用网格测试方法选择最优的惩罚因子C和RBF核的参数〇;调节所述惩罚因子C和参数〇,将识别率最高时对应的参数设为最优参数;其中,惩罚因子C的取值范围可以为[2Λ2 12],所述参数σ的取值范围可以为[2Λ21°];上述训练过程中,由于训练数据是从采集数据中随机抽取的,因此此过程可以重复若干次; [0053] For the SVM classifier training process using the test method to select the optimal grid penalty factor C and RBF kernel parameter square; C and adjusting the penalty factor parameter square, corresponding to the highest recognition rate the parameter is set to the optimum parameters; wherein the penalty factor may be in the range of C [2Λ2 12], the parameter σ can be in the range [2Λ21 °]; training process described above, since the training data is collected from the random data, so this process can be repeated several times;

[0054] (3)利用所述最优参数在训练数据中再运行一次,得到分类器; [0054] (3) the optimal parameters by using the operation in the training data once again, to give classifier;

[0055] (4)利用所述测试数据在该分类器上进行测试准确率,测试完成后得到预设分类器。 [0055] (4) using the test data for testing the accuracy of the classifier, after the test is completed to give a predetermined classifier.

[0056] 由于脑电信号的个人特异性很强,并且脑电信号的强度很弱,在信号采集时,极易被外界信号所干扰。 [0056] Since the highly specific personal EEG, and EEG intensity is weak, when signal acquisition can easily be disturbed by external signals. 因此,事先采集训练数据训练出来的分类器,对于部分测试数据来说其效果并不理想。 Thus, the training data collected previously trained classifier, for some test data which is not ideal.

[0057] 基于上述现象,在本步骤中,设置了清醒状态和睡眠状态的二分类的检测器以对特征数据进行标注,进而通过标注的特征数据训练出符合个人特性的新分类器,以更新预设分类器,取代其用来检测用户的睡眠状态。 [0057] Based on the above phenomenon, in this step, provided the awake state and a sleep state dichotomous detector to annotate the characteristic data, and further training of new classifiers suit individual characteristic by characteristic data annotation, to update default classifier, which substituted for detecting a user's sleep state.

[0058] 上述检测器一般选取一定敏感度(sensitivity)的前提下,具有较高的准确度(precision)的检测器。 Premise [0058] The detector typically select a certain sensitivity (Sensitivity), with high accuracy (precision) of the detector.

[0059] 另外,为了获得较为理想的检测器,第一检测器和第二检测器可以采用理想检测器,利用调整对应样本的惩罚因子的方法来训练所述第一检测器和第二检测器。 [0059] Further, in order to obtain an ideal detector, the first detector and the second detector may be used over the detector, using a method for adjusting the penalty factor corresponding to the samples to train the first detector and the second detector .

[0060] 实验结果表明,这两种检测器的敏感度均高于70%,准确度均高于95%。 [0060] The results show that both the sensitivity of the detector is higher than 70%, higher than 95% accuracy.

[0061] 敏感度指的是所有第i类样本中,被准确识别的比例。 [0061] Sensitivity refers to all class i samples, the ratio is accurately identified. 精确度指的是在所有被识别成第i类的样本中,真实属于第i类的样本比例。 Refers to the accuracy of all the samples is identified as the i-th class, the ratio of the real samples belonging to class i.

[0062] 在一个实施例中,对于预设分类器和检测器1~N的功能可以设置如下: [0062] In one embodiment, the default for the classification and 1 ~ N detector function may be provided as follows:

[0063] 所述预设分类器用于检测用户是否处于1~N种睡眠状态中的任一种,输出结果为"睡眠状态Γ "睡眠状态2"、……或"睡眠状态N" ; [0063] The preset classifier for detecting whether the user is any one of 1 ~ N types of sleep state, the output is "sleep Gamma]" sleep state 2 ", ...... or" sleep N ";

[0064] 所述检测器1用于检测用户是否处于"睡眠状态Γ的状态中,输出结果为"真",反之则输出结果为"假"; [0064] The detector for detecting whether the user is in a "sleep state Γ state, the output is" true ", otherwise the output is" false ";

[0065] 所述检测器2用于检测用户是否处于"睡眠状态2"的状态中,输出结果为"真",反之则输出结果为"假"; [0065] The detector for detecting whether the user 2 is in the "sleep state 2" state, the output is "true", otherwise the output is "false";

[0066] …… [0066] ......

[0067] 所述检测器N用于检测用户是否处于"睡眠状态N"的状态中,输出结果为"真",反之则输出结果为"假"。 Whether the [0067] N detector for detecting a user in a "sleep state N" state, the output is "true", otherwise the output is "false."

[0068] 步骤S103,若只有一个检测器的输出结果为真,则以该检测器的输出结果检测用户的睡眠状态,并对该特征数据进行类型标注;若有多个检测器的输出结果为真,或者全部检测器的输出结果均为假,则以预设分类器的输出结果检测用户的睡眠状态; [0068] step S103, the output result if only one detector is true, places the output of the detector detecting the user's sleep state, and the characteristics of the type of annotation data; output if a plurality of detectors true output, or all of the detectors are false, places a preset sleep detection output of the user classifier;

[0069] 此步骤是基于预设分类器和检测器1~N的识别结果,对用户所处睡眠状态的判断方案。 [0069] This step is based on predetermined classification and a detector 1 to N of the recognition result, the program determines which of the user's sleep state.

[0070] 进一步的,依据以下检测策略进行判断: [0070] Further, detection strategy based on the following judgment:

[0071] ⑴若检测器1~N的输出结果只有一个为"真",其他检测器的输出结果均为"假", 则根据输出结果为"真"的检测器的输出结果检测用户的睡眠状态,将该检测器输出的特征数据的类型标注为相应的类型; [0071] ⑴ When the detector 1 to output only one N is "true", the output of the other detectors are "false", the output is "true" detector for detecting the output according to the user's sleep state, the type of the detected output characteristic data corresponding to the type of labeling;

[0072] 上述方案中,对于特征数据类型的标注后,这些特征数据可以用于训练预设分类器,从而提高分类器的检测睡眠状态的准确性。 [0072] In the above embodiment, the characteristic data for the type of label, such features may be used to train a preset data classifier, thereby improving the detection accuracy of the sleep state classifier.

[0073] (2)若检测器1~N的输出结果中多于一个检测器的输出结果为"真",或检测器1~ N的所有输出结果均为"假",则根据预设分类器的输出结果检测用户的睡眠状态,且不对检测器输出的特征数据进行标注; [0073] (2) If more than one detector 1 to detector output N of the output is "true", or the detector outputs 1 to N are all "false", the default classification output detecting a user's sleep state, and does not feature data detector output is labeled;

[0074] 上述方案中,由于检测器无法检测,因此可以依据预设分类器的检测结果确定当前用户的睡眠状态,此时检测器1~N的输出特征数据不能用于提高预设分类器的训练样本,因此将其丢弃。 [0074] In the above embodiment, since the detection can not detect, and therefore can determine the current sleep state of the user based on a detection result of the preset classifier, when the detector 1 ~ N output characteristic data can not be used to improve preset classifier training sample, and therefore discarded.

[0075] 步骤S104,利用标注的特征数据对所述预设分类器进行训练得到新的分类器,并利用该新的分类器取代所述预设分类器,检测用户的睡眠状态。 [0075] step S104, the feature labeled using the preset data to get a new classifier is trained classifier, and a classifier utilizing the novel substituted the preset classifier, detecting a user's sleep state.

[0076] 在此步骤中,基于前述步骤S103中已标注的特征数据,将其作为样本输入到预设分类器中进行训练得到新的分类器,用这个新的分类器取代预设分类器,从而可以提高预设分类器的检测睡眠状态准确性。 [0076] In this step, based on the feature data of the already marked in step S103, the preset input to train the classifier to obtain new classifier, with this new classifier classifier substituted preset as a sample, thereby improving the pre-classifier to detect sleep accuracy.

[0077] 在实际应用中,随着用户的不断使用,可以持续触发,不断更新分类器,从而可以不断准确性,而且当应用到其他用户时,也可以重新训练出分类器,得到更适合该用户的分类器。 [0077] In practical applications, with the user, you can keep firing constantly updated classifier, which can continue to accuracy, and when applied to other users, can also be re-trained classifier to be more appropriate for the user classifier.

[0078] 在一个实施例中,在训练新的分类器时,首先判断已标注类型的特征数据的数量, 当数量达到设定阈值时,将已标注的特征数据作为样本数据输入预设分类器进行训练得到新的分类器; [0078] In one embodiment, in training new classifier first determines the type of the labeled amount of characteristic data, when the number reaches a set threshold value, characteristic data which has been labeled as the default input sample data classifier training to get new classifier;

[0079] 通过以设定阈值,当收集到的标注类型的特征数据达到一定数量时,输入预设分类器进行训练,避免样本数量过低,训练效果不佳。 [0079] By setting a threshold value, when the type of annotation is collected a certain number of feature data, the default input classifier is trained, the number of samples to avoid too low, poor training effect.

[0080] 参考图3所示,图3是多分类器融合检测器检测睡眠状态的示意图。 [0080] Referring to FIG. 3, FIG. 3 is a schematic view of a multi-classifier fusion sleep state detector. 标注过程中除了利用其他样本数据的较为平衡的预设分类器之外,还设计N个检测器,检测器1用于检测用户是否处于"睡眠状态1"、检测器2用于检测用户是否处于"睡眠状态2"的状态中,……, 检测器N用于检测用户是否处于"睡眠状态N"的状态中。 Labeling process more balanced except for using preset classification other than the sample data, designed N detectors, a detector for detecting whether the user is in a "sleep state 1", the detector 2 for detecting whether or not the user is status "sleep state 2" in, ......, N detector for detecting whether the user is in the "sleep state N" state.

[0081] 特征数据输入后分别进入预设分类器和检测器1~N;通过上述检测策略进行判断,输入用户当前的睡眠状态检测结果,对于标注数据类型的特征数据,输入至预设分类器进行训练新的分类器,对于未标注数据类型的特征数据,检测后将其丢弃。 [0081] After the feature data are input into the preset classification and detector 1 ~ N; judgment, the user inputs the current sleep state detection result by the detection strategy, for the feature data annotation data type, the default input to the classifier new classifier training, the feature data for the data type of unlabeled, detection will be discarded.

[0082] 本发明的基于多分类器融合的睡眠状态检测方法可以检测各种睡眠状态,例如, 用来检测非眼快动睡眠的状态时,可以检测入睡期,浅睡期,中等睡眠期和深度睡眠期等四个状态,分别对应于检测器1、检测器2、检测器3和检测器4。 [0082] The present invention can detect various sleep state based on the sleep state detecting method for fusion of multiple classifiers, e.g., for detecting when a non-rapid eye movement sleep state can be detected in sleep, light sleep, and sleep medium four deep sleep state, corresponding to detector 1, detector 2, the detector 3 and the detector 4.

[0083] 参考图4所示,图4为一个实施例的基于多分类器融合的睡眠状态检测系统结构示意图,包括: [0083] With reference to FIG, 4 is a schematic structural diagram of a sleep state detecting FIG 4 embodiment is based on a multi-classifier fusion system, comprising:

[0084] 特征数据提取模块101,用于采集用户在睡眠过程中产生的脑电信号,根据睡眠状态识别的识别任务从所述脑电信号中提取相应的特征数据; [0084] The feature data extraction module 101, a user generated acquisition EEG during sleep, the recognition task identification feature data extracted from the respective EEG The sleep state;

[0085] 分类器融合检测模块102,用于将所述特征数据分别输入预设分类器、检测器1~N 进行检测;其中,所述预设分类器检测用户的1~N种睡眠状态,检测器1~N分别对应检测用户的一种特定睡眠状态,N彡2; [0085] classifier fusion detection module 102, the feature data for each predetermined input classifier, 1 ~ N detectors for detecting; wherein said predetermined sleep state 1 ~ N kinds classifier detects a user, 1 ~ N detectors respectively detecting a particular user's sleep state, N San 2;

[0086] 结果判断和数据标注模块103,用于若只有一个检测器的输出结果为真,则以该检测器的输出结果检测用户的睡眠状态,并对该特征数据进行类型标注;若有多个检测器的输出结果为真,或者全部检测器的输出结果均为假,则以预设分类器的输出结果检测用户的睡眠状态; [0086] Analyzing the results of the data and the annotation module 103, if only for a detector output is true, places the output of the detector detecting the user's sleep state, and the type of the annotation feature data; if multiple the output of the detectors is true, or the output of the detector are all false, the user places a predetermined detection output of classifier sleep state;

[0087] 分类器训练和更新模块103,用于利用标注的特征数据对所述预设分类器进行训练得到新的分类器,并利用该新的分类器取代所述预设分类器,检测用户的睡眠状态。 [0087] classifier training and updating module 103, using the feature data for the marked preset classifier is trained classifier to obtain a new, and utilizing the new classifier classifier instead of the preset, the user detection sleep state.

[0088] 本发明的基于多分类器融合的睡眠状态检测系统与本发明的基于多分类器融合的睡眠状态检测方法一一对应,在上述基于多分类器融合的睡眠状态检测方法的实施例阐述的技术特征及其有益效果均适用于基于多分类器融合的睡眠状态检测系统的实施例中, 特此声明。 [0088] Based on the sleep state detection method based on multiple classifiers Fusion Fusion with multiple one-sleep state detection system according to the present invention, the present invention is set forth in the above-described embodiment is based on the sleep state detection method fused multiple classifiers the technical features and advantageous effects are applicable to the embodiment of the multi fusion with detection of sleep state based system, it is hereby declared.

[0089] 以上所述实施例的各技术特征可以进行任意的组合,为使描述简洁,未对上述实施例中的各个技术特征所有可能的组合都进行描述,然而,只要这些技术特征的组合不存在矛盾,都应当认为是本说明书记载的范围。 [0089] each of the above embodiments of the technical features of any combination can be, for the brevity, not each of the technical features of the embodiments described above are all the possible combinations will be described, however, as long as the combination of these features is not contradiction, they are to be considered in the scope described in this specification.

[0090] 以上所述实施例仅表达了本发明的几种实施方式,其描述较为具体和详细,但并不能因此而理解为对发明专利范围的限制。 [0090] The above embodiments are only expressed several embodiments of the present invention, and detailed description thereof is more specific, but can not therefore be understood to limit the scope of the invention. 应当指出的是,对于本领域的普通技术人员来说,在不脱离本发明构思的前提下,还可以做出若干变形和改进,这些都属于本发明的保护范围。 It should be noted that those of ordinary skill in the art, without departing from the spirit of the present invention, can make various changes and modifications, which fall within the protection scope of the present invention. 因此,本发明专利的保护范围应以所附权利要求为准。 Therefore, the protection scope of the present invention should be subject to the appended claims.

Claims (10)

1. 一种基于多分类器融合的睡眠状态检测方法,其特征在于,包括: 采集用户在睡眠过程中产生的脑电信号,根据睡眠状态识别的识别任务从所述脑电信号中提取相应的特征数据; 将所述特征数据分别输入预设分类器、检测器1~N进行检测;其中,所述预设分类器检测用户的1~N种睡眠状态,检测器1~N分别对应检测用户的一种特定睡眠状态,N多2; 若只有一个检测器的输出结果为真,则以该检测器的输出结果检测用户的睡眠状态, 并对该特征数据进行类型标注;若有多个检测器的输出结果为真,或者全部检测器的输出结果均为假,则以预设分类器的输出结果检测用户的睡眠状态; 利用标注的特征数据对所述预设分类器进行训练得到新的分类器,并利用该新的分类器取代所述预设分类器,检测用户的睡眠状态。 CLAIMS 1. A method for detecting a sleep state based on the fusion of multiple classifiers, characterized in that, comprising: a user generated acquisition EEG during sleep, the recognition task identification is extracted from the EEG sleep state according to the respective feature data; wherein said preset data are inputted classifier, 1 ~ N detectors for detecting; wherein said predetermined sleep state 1 ~ N kinds classifier detects a user, the detector 1 ~ N respectively detect user one particular sleep state, N number 2; if only one detector output is true, places the output of the detector detecting the user's sleep state, and the type of the annotation feature data; if a plurality of detection the output unit is true, or the output of the detector are all false, the user places a predetermined detection output of classifier sleep; labeled using the feature data obtained for the new preset classifier training classifier, and a classifier utilizing the novel substituted the preset classifier, detecting a user's sleep state.
2. 根据权利要求1所述的基于脑电信号的睡眠状态识别模型训练方法,其特征在于,所述预设分类器用于检测用户是否处于1~N种睡眠状态中的任一种,输出结果为"睡眠状态1" "睡眠状态2"、……或"睡眠状态N" ; 所述检测器1用于检测用户是否处于"睡眠状态1"的状态中,输出结果为"真",反之则输出结果为"假"; 所述检测器2用于检测用户是否处于"睡眠状态2"的状态中,输出结果为"真",反之则输出结果为"假"; 所述检测器N用于检测用户是否处于"睡眠状态N"的状态中,输出结果为"真",反之则输出结果为"假"。 The sleep EEG recognition model training method based on claim 1 wherein the preset claim classifier for detecting whether the user is any one kind of 1 ~ N in the sleep state, the output to "sleep state 1" "sleep state 2", ...... or "sleep N"; the detector for detecting whether the user is in a "sleep state 1" state, the output is "true", otherwise output is "false"; if the detector 2 for detecting a user in a state of "sleep state 2", the output is "true", otherwise the output is "false"; the detector used N detecting whether the user is in the "sleep state N" state, the output is "true", otherwise the output is "false."
3. 根据权利要求1所述的基于脑电信号的睡眠状态识别模型训练方法,其特征在于,若检测器1~N的输出结果只有一个为"真",其他检测器的输出结果均为"假",则根据输出结果为"真"的检测器的输出结果检测用户的睡眠状态,将该检测器输出的特征数据的类型标注为相应的类型。 3. Training Method sleep state recognition model based on EEG, wherein according to claim 1, when the output of the detector 1 to N, only one of "true", the output of the other detectors are " false ", the output result is" detection output of the user's sleep state true "detector, the detector output type of feature data corresponding to the type of labeling.
4. 根据权利要求1所述的基于脑电信号的睡眠状态识别模型训练方法,其特征在于,若检测器1~N的输出结果中多于一个检测器的输出结果为"真",或检测器1~N的所有输出结果均为"假",则根据预设分类器的输出结果检测用户的睡眠状态,且不对检测器输出的特征数据进行标注。 4. sleep state recognition model training method based on EEG, wherein according to claim 1, if more than one detector 1 to detector output N of the output is "true", or detection 1 ~ N are all output, "false", according to a preset classification result of the detection output of the user's sleep state, and does not feature data detector output is labeled.
5. 根据权利要求1所述的基于多分类器融合的睡眠状态检测方法,其特征在于,利用标注的特征数据对所述预设分类器进行训练得到新的分类器的步骤包括: 判断已标注类型的特征数据的数量,当数量达到设定阈值时,将已标注的特征数据作为样本数据输入预设分类器进行训练得到新的分类器。 The method for detecting a sleep state based on multiple classifiers fusion, wherein the 1, labeled using the preset characteristic data of the classifier is trained classifier to obtain a new step comprises: determining the labeled number of types of characteristic data, when the number reaches a set threshold value, characteristic data which has been labeled as the default input sample data to obtain a new classifier is trained classifier.
6. 根据权利要求1所述的基于多分类器融合的睡眠状态检测方法,其特征在于,所述从所述脑电信号中提取相应的特征数据的步骤包括: 提取脑电信号的基线,计算所述基线的变化幅度;其中,所述变化幅度为基线最大值减去最小值; 在去掉基线后,对所述脑电信号进行小波分解,获得小波系数,并根据小波系数计算小波系数的特征参数;其中,所述特征参数包括小波系数的均值、方差、峭度系数和/或斜度系数; 在去掉基线后,计算脑电信号的LZ复杂度和样本熵; 将所述基线的变化幅度、小波系数的特征参数、LZ复杂度和样本熵设为所述特征数据。 6. A method for detecting a sleep state based on the fusion of multiple classifiers, characterized in that according to claim 1, wherein the step corresponding to said data comprises extracting from said EEG signals: extracting baseline EEG was calculated the baseline variation width; wherein said change in amplitude of the maximum minus the minimum baseline; baseline after removal, the EEG wavelet decomposition to obtain wavelet coefficients, wavelet coefficients and wavelet coefficients is calculated according to the features parameters; wherein said characteristic parameters comprise wavelet coefficients mean, variance, kurtosis coefficients and / or slope coefficient; after removing the baseline EEG LZ calculation complexity and sample entropy; the variation width of the baseline characteristic parameters of the wavelet coefficients, LZ sample entropy and complexity to the feature data.
7. 根据权利要求6所述的基于多分类器融合的睡眠状态检测方法,其特征在于,还包括: 在小波重构中提取所述脑电信号S波频段、0波频段、a波频段和财皮频段的信号; 分别计算脑电信号中S波频段,0波频段,a波频段4波频段的能量在总能量中的比例; 分别计算在一帧的时间内,脑电信号中S波频段,0波频段,a波频段,財皮频段能量最大的时间长度; 将所述比例和时间设为所述特征数据。 7. A method for detecting a sleep state based on the fusion of multiple classifiers, characterized in that said claim 6, further comprising: extracting the S-wave EEG wavelet reconstruction frequency band, the wave band 0, A wave band and Choi leather band signal; EEG S-wave band, the wave band 0, band wave energy band 4 in a ratio of a total energy waves are calculated; values ​​are calculated in the time of one frame, the S-wave EEG band, 0-wave band, the wave band a, the maximum length of time financial transdermal band energy; and the time ratio of the characteristic data set.
8. 根据权利要求7所述的基于多分类器融合的睡眠状态检测方法,其特征在于,所述分别计算脑电信号中S波频段,0波频段,a波频段,財皮频段的能量在总能量中的比例的步骤包括如下公式: rs 一S (ys) /ptotai 1*9 一S (yfl) /ptotai Ta- Xl (y〇) /ptotai rp- S (yp) /ptotai 其中口1;。 According to claim 7 sleep state detection method based on multiple classifiers fusion, wherein said EEG signals were calculated S-wave band, the wave band 0, A wave band, Choi leather band energy the proportion of the total energy step comprises the following formula: rs a S (ys) / ptotai 1 * 9 a S (yfl) / ptotai Ta- Xl (y〇) / ptotai rp- S (yp) / ptotai port 1 wherein; . 1;£11=乙(75)2+乙(7(3)2+乙(7 (1)2+2(7{!)2,75,7(3,7(1和7£!分别表示重构后的5频段、0 频段、a频段和13频段的信号,rS,re,ra和re分别代表S频段、0频段、a频段和仏频段的信号的能量在总能量的比例。 1;!! £ 11 = B (75) B 2 + (7 (3) 2 + B (7 (1) 2 + 2 ({7) 2,75,7 (3, 7 (1 and 7 represent £ 5-band reconstructed, band 0, band 13 and a band signal, rS, re, ra and re represent the S-band, the frequency ratio of the total energy, energy of the signal and a frequency band of 0 Fo.
9. 根据权利要求7所述的基于多分类器融合的睡眠状态检测方法,其特征在于,所述分别计算在一帧的时间内,脑电信号中S波频段,0波频段,a波频段,財皮频段能量最大的时间长度的步骤包括如下公式: 9. A method for detecting a sleep state based on the fusion of multiple classifiers, characterized in that according to claim 7, the values ​​are calculated in the time of one frame, the S-wave EEG frequency band, the wave band 0, band A wave , the maximum step time length financial transdermal band energy comprises the following formula:
Figure CN106333680AC00031
式中,CS,CO,Ca和Cf!表示S频段、0频段、a频段和0频段的信号在当前帧内所占能量比例最大的时间长度,'分别表示第i秒内S频段、0频段、a频段和0频段的信号的能量在总能量的比例。 Wherein, CS, CO, Ca and of Cf! Represents S-band, the frequency band 0, band A and band signal frame 0 of the maximum length of time the proportion of energy in the current, 'denote the i-th second S-band, the frequency band 0 , a proportion of the total energy and the energy of the signal band 0 to band.
10. -种基于多分类器融合的睡眠状态检测系统,其特征在于,包括: 特征数据提取模块,用于采集用户在睡眠过程中产生的脑电信号,根据睡眠状态识别的识别任务从所述脑电信号中提取相应的特征数据; 分类器融合检测模块,用于将所述特征数据分别输入预设分类器、检测器1~N进行检测;其中,所述预设分类器检测用户的1~N种睡眠状态,检测器1~N分别对应检测用户的一种特定睡眠状态,N彡2; 结果判断和数据标注模块,用于若只有一个检测器的输出结果为真,则以该检测器的输出结果检测用户的睡眠状态,并对该特征数据进行类型标注;若有多个检测器的输出结果为真,或者全部检测器的输出结果均为假,则以预设分类器的输出结果检测用户的睡眠状态; 分类器训练和更新模块,用于利用标注的特征数据对所述预设分类器进行训练得到新的分类器 10 - Multi-Species Fusion with detection system based on the sleep state, characterized by comprising: feature data extraction means for collecting EEG signals generated by the user during sleep, sleep state in accordance with recognition task identified from the EEG extracting corresponding feature data; classifier fusion detection module configured to preset the input feature data are classified, detector detects 1 ~ N; wherein said predetermined classification detecting user 1 ~ N kinds of sleep, 1 ~ N detectors respectively detecting a particular user's sleep state, N San 2; and the result of the determination data labeling module, if only for a detector output is true, places the detection output detecting user's sleep state, and the type of the annotation feature data; if a plurality of detector output is true, or the output of the detector are all false, the output of the classifier preset places results detects the user's sleep state; classifier training and updating module, for utilizing feature data of the predetermined label classifier is trained classifier to obtain a new 并利用该新的分类器取代所述预设分类器,检测用户的睡眠状态。 And utilizing the new classifier substituted by the predetermined classification which detects a user's sleep state.
CN201610843526.5A 2016-09-21 2016-09-21 Sleep state detection method and system based on fusion of multiple classifiers CN106333680A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201610843526.5A CN106333680A (en) 2016-09-21 2016-09-21 Sleep state detection method and system based on fusion of multiple classifiers

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201610843526.5A CN106333680A (en) 2016-09-21 2016-09-21 Sleep state detection method and system based on fusion of multiple classifiers

Publications (1)

Publication Number Publication Date
CN106333680A true CN106333680A (en) 2017-01-18

Family

ID=57839170

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201610843526.5A CN106333680A (en) 2016-09-21 2016-09-21 Sleep state detection method and system based on fusion of multiple classifiers

Country Status (1)

Country Link
CN (1) CN106333680A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107708191A (en) * 2017-09-30 2018-02-16 北京小米移动软件有限公司 Terminal control method and apparatus, and computer readable storage medium

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20090203972A1 (en) * 2006-06-01 2009-08-13 Biancamed Ltd. Apparatus, system, and method for monitoring physiological signs
CN101515200A (en) * 2009-04-03 2009-08-26 北京工业大学 Target selecting method based on transient visual evoked electroencephalogram
CN104573630A (en) * 2014-12-05 2015-04-29 杭州电子科技大学 Multiclass brain electrical mode online identification method based on probability output of twin support vector machine
US20160022201A1 (en) * 2014-09-23 2016-01-28 Fitbit, Inc. Automatic change of power consumption of sensors responsive to user's state transition
CN105286890A (en) * 2015-09-22 2016-02-03 江西科技学院 Driver sleepy state monitoring method based on electroencephalogram signal
CN105792733A (en) * 2013-11-28 2016-07-20 皇家飞利浦有限公司 Sleep monitoring device

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20090203972A1 (en) * 2006-06-01 2009-08-13 Biancamed Ltd. Apparatus, system, and method for monitoring physiological signs
CN101515200A (en) * 2009-04-03 2009-08-26 北京工业大学 Target selecting method based on transient visual evoked electroencephalogram
CN105792733A (en) * 2013-11-28 2016-07-20 皇家飞利浦有限公司 Sleep monitoring device
US20160022201A1 (en) * 2014-09-23 2016-01-28 Fitbit, Inc. Automatic change of power consumption of sensors responsive to user's state transition
CN104573630A (en) * 2014-12-05 2015-04-29 杭州电子科技大学 Multiclass brain electrical mode online identification method based on probability output of twin support vector machine
CN105286890A (en) * 2015-09-22 2016-02-03 江西科技学院 Driver sleepy state monitoring method based on electroencephalogram signal

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107708191A (en) * 2017-09-30 2018-02-16 北京小米移动软件有限公司 Terminal control method and apparatus, and computer readable storage medium

Similar Documents

Publication Publication Date Title
Kiymik et al. Automatic recognition of alertness level by using wavelet transform and artificial neural network
Temko et al. EEG-based neonatal seizure detection with support vector machines
Ocak Optimal classification of epileptic seizures in EEG using wavelet analysis and genetic algorithm
Indiradevi et al. A multi-level wavelet approach for automatic detection of epileptic spikes in the electroencephalogram
Liu et al. Automatic seizure detection using wavelet transform and SVM in long-term intracranial EEG
Nenadic et al. Spike detection using the continuous wavelet transform
Motamedi-Fakhr et al. Signal processing techniques applied to human sleep EEG signals—A review
Halford Computerized epileptiform transient detection in the scalp electroencephalogram: Obstacles to progress and the example of computerized ECG interpretation
CN101690659A (en) Brain wave analysis method
Ahammad et al. Detection of epileptic seizure event and onset using EEG
US20030105408A1 (en) Artifact detection in encephalogram data using an event model
CN102274032A (en) Driver fatigue detection system based on EEG
Mardi et al. EEG-based drowsiness detection for safe driving using chaotic features and statistical tests
CN103412646B (en) Music brain-computer interface based on the recommended method of emotional
Sharma et al. An automatic detection of focal EEG signals using new class of time–frequency localized orthogonal wavelet filter banks
Tiwari et al. Automated diagnosis of epilepsy using key-point-based local binary pattern of EEG signals
CN101596101B (en) Method for determining fatigue state according to electroencephalogram
Taylor et al. Automatic identification of artifacts in electrodermal activity data
Radüntz et al. EEG artifact elimination by extraction of ICA-component features using image processing algorithms
Majumdar et al. Automatic seizure detection in ECoG by differential operator and windowed variance
Patidar et al. Detection of epileptic seizure using Kraskov entropy applied on tunable-Q wavelet transform of EEG signals
Zhang et al. Recognition of mental workload levels under complex human–machine collaboration by using physiological features and adaptive support vector machines
CN101219048B (en) Method for extracting brain electrical character of imagine movement of single side podosoma
Khan et al. Automatic detection of seizure onset in pediatric EEG
Hsu Assembling a multi-feature EEG classifier for left–right motor imagery data using wavelet-based fuzzy approximate entropy for improved accuracy

Legal Events

Date Code Title Description
C06 Publication
C10 Entry into substantive examination