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

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

Sleep state detection method based on multiple Classifiers Combination and system

Technical field

The present invention relates to assisting sleep technical field, more particularly to a kind of sleep state inspection based on multiple Classifiers Combination Survey method and system.

Background technology

In sleep, human body has carried out the process self loosened and recover, and therefore good sleep is to maintain healthy A primary condition；But due to the reason such as operating pressure is big, daily life system is irregular, result in the sleep matter of part population Amount is not good enough, shows as insomnia, midnight wakes up with a start.

There are some equipment at present on the market to help people to fall asleep, improved sleep quality.For example specific sleep a certain Pass through the manual intervention such as sound, optical signal, it is to avoid wake user etc. under the state of sleeping soundly under dormancy state.Assisting sleep is set For standby, in order to be really achieved the purpose improving user's sleep quality, the sleep state of correct detection user is extremely important 's.

Clinically mainly adopt at present polysomnogram to identify sleep state, mainly use EEG signals come to sleep into Row analysis, identifies the sleep state of measured by training sleep state model, for example, judges which of sleep user be in Stage, but because the individual human specific of EEG signals is very strong, and intensity is very weak, in signals collecting easily by outer signals institute Interference.There is error to the detection of a lot of users in therefore general training in advance grader out, accuracy is difficult to be guaranteed.

Content of the invention

Based on this it is necessary to be directed to the problems referred to above, provide a kind of sleep state detection method based on multiple Classifiers Combination And system, effectively improve the accuracy of default grader identification.

A kind of sleep state detection method based on multiple Classifiers Combination, comprising:

The EEG signals that collection user produces in sleep procedure, the identification mission according to sleep state identification is from described brain Corresponding characteristic is extracted in the signal of telecommunication；

Described characteristic is inputted default grader respectively, detector 1～n is detected；Wherein, described default classification Device detects 1～n kind sleep state of user, a kind of detector 1～n corresponding specific sleep state detecting user respectively, n >=2；

The sleep shape of user if the output result of only one of which detector is true, is detected with the output result of this detector State, and type mark is carried out to this feature data；If the output result having multiple detectors is true, or whole detector is defeated Go out result and be vacation, then detect the sleep state of user with the output result of default grader；

Characteristic using mark is trained obtaining new grader to described default grader, and new using this Grader replaces described default grader, the sleep state of detection user.

A kind of sleep state detecting system based on multiple Classifiers Combination, comprising:

Characteristic extraction module, for gathering the EEG signals that user produces in sleep procedure, according to sleep state The identification mission of identification extracts corresponding characteristic from described EEG signals；

Multiple Classifier Fusion detection module, for described characteristic is inputted default grader respectively, detector 1～n enters Row detection；Wherein, 1～n kind sleep state of described default detection of classifier user, detector 1～n corresponding detection user respectively A kind of specific sleep state, n >=2；

Result judges data labeling module, if the output result for only one of which detector is true, with this detection The output result of device detects the sleep state of user, and carries out type mark to this feature data；If there being the defeated of multiple detectors It is true for going out result, or the output result of whole detector is vacation, then detect user's with the output result of default grader Sleep state；

Classifier training and update module, for being trained to described default grader using the characteristic of mark To new grader, and replace described default grader, the sleep state of detection user using this new grader.

The above-mentioned sleep state detection method based on multiple Classifiers Combination and system, based on the characteristic of EEG signals, On the basis of default grader, it is provided with further for multiple dormant detector 1～n, integrated classification device detection knot Fruit and multiple detector output results carry out type mark to characteristic, are then inputted pre- by the characteristic of marking types If classifier training goes out new grader, replace former default grader, the sleep state of detection user.The program can use During train and be closer to the specific grader of individual subscriber, the accuracy rate of grader can be improved, strengthen sleep The accuracy of state-detection.

Brief description

Fig. 1 is the flow chart of the sleep state detection method based on multiple Classifiers Combination of an embodiment；

Fig. 2 is the EEG signals schematic diagram before and after Filtering Processing；

Fig. 3 is that multiple Classifiers Combination detector detects dormant schematic diagram；

Fig. 4 is the sleep state detecting system structural representation based on multiple Classifiers Combination of an embodiment.

Specific embodiment

Below in conjunction with the accompanying drawings illustrate the present invention based on the sleep state detection method of multiple Classifiers Combination and the reality of system Apply example.

With reference to shown in Fig. 1, Fig. 1 is the flow process of the sleep state detection method based on multiple Classifiers Combination of an embodiment Figure, comprising:

Step s101, the EEG signals that collection user produces in sleep procedure, appointed according to the identification of sleep state identification Corresponding characteristic is extracted in business from described EEG signals；

In this step, as when assisting sleep is carried out to user, related transducer equipment is worn by user, detect user EEG signals, gather EEG signals when, can be acquired with 30s for a frame.

Carry out the task of sleep state identification as needed, determine feature data types, extract therewith from EEG signals Corresponding characteristic；For example, 1～n kind sleep state to be identified, extract the characteristic for carrying out this n kind state recognition.

In one embodiment, before extracting characteristic, the EEG signals that be gathered can also be filtered process, filter Except high-frequency noise and Hz noise.For example, the useful information of EEG signals focuses mostly in the range of 0-100hz, is gathering Cheng Zhonghui mixes frequency in this extraneous noise, therefore, it can be filtered by means of filtering.Can same band filter Filter high-frequency noise, and design a wave trap (50/60hz) to filter Hz noise.

With reference to shown in Fig. 2, Fig. 2 is the EEG signals schematic diagram before and after Filtering Processing, and upper figure is primary signal, figure below be through Cross the signal after Filtering Processing, it can be found that most high-frequency noise is filtered out.

For the scheme extracting characteristic, the present invention provides some embodiments, and detailed process includes the following:

(1) extract the baseline of EEG signals, calculate the amplitude of variation of described baseline；Wherein, described amplitude of variation is baseline Maximum deducts minima；

(2) after removing baseline, described EEG signals are carried out with wavelet decomposition, obtain wavelet coefficient, and according to wavelet systems Number calculates the characteristic parameter of wavelet coefficient；Wherein, described characteristic parameter include the average of wavelet coefficient, variance, kurtosis coefficient and/ Or gradient coefficient；

In order to preferably decomposite described various frequency waveform, the number of plies of wavelet decomposition is full with the sample frequency of EEG signals Following relation: the f=2 of footⁿ⁺², wherein, f is the sample frequency of EEG signals, and n is the number of plies of wavelet decomposition；For example, when signal When down-sampled rate is 128hz, 4 layers of decomposition can be selected, when the sample rate of signal is 256hz, then can carry out 5 layers of decomposition.

(3) lz complexity and the Sample Entropy of EEG signals after removing baseline, are calculated；

The amplitude of variation of described baseline, the characteristic parameter of wavelet coefficient, lz complexity and Sample Entropy are set to described feature Data；

By the scheme of above-described embodiment, the data as signal characteristic includes the amplitude of variation of baseline, wavelet coefficient Characteristic parameter, lz complexity and Sample Entropy etc..

Further, can also be identified using the waveform of multiple wave bands of EEG signals, carry in wavelet reconstruction Take described EEG signals δ wave frequency section, the signal of θ wave frequency section, α wave frequency section and β wave frequency section；According to the difference of frequency, EEG signals It is to be divided into 4 species rhythm brain waves: δ ripple (1-3hz), θ ripple (4-7hz), α ripple (8-12hz), β ripple (14-30hz), here, After the signal of these four frequency ranges can be extracted, calculate correlated characteristic using these signals, concrete scheme can be such that

(4) δ wave frequency section in EEG signals, θ wave frequency section, α wave frequency section are calculated respectively, the energy of β wave frequency section is in gross energy Ratio；This ratio is also served as characteristic input grader be identified；Computational methods can include equation below:

r_δ=∑ (y_δ)²/p_total

r_θ=∑ (y_θ)²/p_total

r_α=σ (y_α)²/p_total

r_β=∑ (y_β)²/p_total

Wherein p_total=∑ (y_δ)²+∑(y_θ)²+∑(y_α)²+σ(y_β)², y_δ, y_θ, y_αAnd y_βRepresent the δ frequency after reconstruct respectively Section, the signal of θ frequency range, α frequency range and β frequency range, r_δ, r_θ, r_αAnd r_βRepresent δ frequency range, the signal of θ frequency range, α frequency range and β frequency range respectively Energy gross energy ratio.

(5) calculate respectively within the time of a frame, δ wave frequency section in EEG signals, θ wave frequency section, α wave frequency section, β wave frequency section energy The maximum time span of amount；This time also served as characteristic input grader be identified, computational methods can include as Lower formula:

$c_{\mathrm{\δ}}=\underset{i=1}{\overset{30}{\σ}}{f}_{\mathrm{\δ}}^i,f_{\mathrm{\δ}}^i=\left\{\begin{array}{ccc}1,& if& r_{\mathrm{\δ}}^i=\max (r_{\mathrm{\δ}}^i,r_{\mathrm{\θ}}^i,r_{\mathrm{\α}}^i,r_{\mathrm{\β}}^i)\\ 0,& if& r_{\mathrm{\δ}}^i\&notequal;\max (r_{\mathrm{\δ}}^i,r_{\mathrm{\θ}}^i,r_{\mathrm{\α}}^i,r_{\mathrm{\β}}^i)\end{array}\right.$

$c_{\mathrm{\θ}}=\underset{i=1}{\overset{30}{\σ}}{f}_{\mathrm{\θ}}^i,f_{\mathrm{\θ}}^i=\left\{\begin{array}{ccc}1,& if& r_{\mathrm{\θ}}^i=\max (r_{\mathrm{\δ}}^i,r_{\mathrm{\θ}}^i,r_{\mathrm{\α}}^i,r_{\mathrm{\β}}^i)\\ 0,& if& r_{\mathrm{\θ}}^i\&notequal;\max (r_{\mathrm{\δ}}^i,r_{\mathrm{\θ}}^i,r_{\mathrm{\α}}^i,r_{\mathrm{\β}}^i)\end{array}\right.$

$c_{\mathrm{\α}}=\underset{i=1}{\overset{30}{\σ}}{f}_{\mathrm{\α}}^i,f_{\mathrm{\α}}^i=\left\{\begin{array}{ccc}1,& if& r_{\mathrm{\α}}^i=\max (r_{\mathrm{\δ}}^i,r_{\mathrm{\θ}}^i,r_{\mathrm{\α}}^i,r_{\mathrm{\β}}^i)\\ 0,& if& r_{\mathrm{\α}}^i\&notequal;\max (r_{\mathrm{\δ}}^i,r_{\mathrm{\θ}}^i,r_{\mathrm{\α}}^i,r_{\mathrm{\β}}^i)\end{array}\right.$

$c_{\mathrm{\β}}=\underset{i=1}{\overset{30}{\σ}}{f}_{\mathrm{\β}}^i,f_{\mathrm{\β}}^i=\left\{\begin{array}{ccc}1,& if& r_{\mathrm{\β}}^i=\max (r_{\mathrm{\δ}}^i,r_{\mathrm{\θ}}^i,r_{\mathrm{\α}}^i,r_{\mathrm{\β}}^i)\\ 0,& if& r_{\mathrm{\β}}^i\&notequal;\max (r_{\mathrm{\δ}}^i,r_{\mathrm{\θ}}^i,r_{\mathrm{\α}}^i,r_{\mathrm{\β}}^i)\end{array}\right.$

In formula, c_δ, c_θ, c_αAnd c_βRepresent δ frequency range, the signal of θ frequency range, α frequency range and β frequency range in energy shared by current frame in The maximum time span of ratio,Represent δ frequency range, the energy of the signal of θ frequency range, α frequency range and β frequency range in i-th second respectively Amount is in the ratio of gross energy.

Step s102, described characteristic is inputted default grader respectively, detector 1～n is detected；Wherein, institute State 1～n kind sleep state of default detection of classifier user, a kind of detector 1～n corresponding specific sleep detecting user respectively State, n >=2；

For above-mentioned default grader, can be using rbf core svm (support vector machin, support to Amount machine) grader, it would however also be possible to employ neutral net, the grader of decision tree.This grader is to be trained by other sample datas Obtain.

Training process can be such that

(1) obtain the characteristic of described user, randomly draw equal number from two kinds of characteristic respectively Sample as training data, remaining is as test data；

(2) described training data is inputted support vector machine classifier or neutral net carries out self study, using grid- Test algorithm finds discrimination highest parameter, and this parameter is set to optimized parameter；

For support vector machine classifier, adopt in training process the optimum penalty factor c of grid software test method choice and Parameter σ of rbf core；Adjust described penalty factor c and parameter σ, corresponding parameter during discrimination highest is set to optimized parameter；Its In, the span of penalty factor c can be [2^-2, 2¹²], the span of described parameter σ can be [2^-2, 2¹⁰]；Above-mentioned instruction During white silk, because training data is randomly drawed from gathered data, therefore this process can be with repeated several times；

(3) reruned in training data once using described optimized parameter, obtain grader；

(4) carry out test accuracy rate using described test data on this grader, after the completion of test, obtain default classification Device.

Because the individual human specific of EEG signals is very strong, and the intensity of EEG signals is very weak, in signals collecting, easily Disturbed by outer signals.Therefore, the grader that in advance collection training data trains out, for partial test data its Effect is unsatisfactory.

Based on above-mentioned phenomenon, in this step, it is provided with waking state and dormant two detectors classified with right Characteristic is labeled, and then trains, by the characteristic of mark, the new grader meeting personal characteristics, pre- to update If grader, replace it for the sleep state detecting user.

On the premise of above-mentioned detector typically chooses certain sensitivity (sensitivity), there is higher accuracy (precision) detector.

In addition, in order to obtain ideal detector, the first detector and the second detector can be using preferable detections Device, the method using the penalty factor of the corresponding sample of adjustment trains described first detector and the second detector.

Test result indicate that, the sensitivity of both detectors is above 70%, and accuracy is above 95%.

Sensitivity refers to the ratio in all i-th class samples, being accurately identified.Degree of accuracy refers to all identified Become the sample proportion truly belonging to the i-th class in the sample of the i-th class.

In one embodiment, the function for default grader and detector 1～n can be provided that

Described default grader is used for detecting any one whether user be in 1～n kind sleep state, output result is " sleep state 1 " " sleep state 2 " ... or " sleep state n "；

Described detector 1 is used for detecting that in the state that whether user is in " sleep state 1 ", output result is " very ", instead Then output result be " vacation "；

Described detector 2 is used for detecting that in the state that whether user is in " sleep state 2 ", output result is " very ", instead Then output result be " vacation "；

……

Described detector n is used for detecting that in the state that whether user is in " sleep state n ", output result is " very ", instead Then output result be " vacation ".

Step s103, if the output result of only one of which detector is true, is used with the output result detection of this detector The sleep state at family, and type mark is carried out to this feature data；If the output result having multiple detectors is true, or all The output result of detector is vacation, then detect the sleep state of user with the output result of default grader；

This step is the recognition result based on default grader and detector 1～n, to judgement dormant residing for user Scheme.

Further, judged according to following inspection policies:

(1) if the output result only one of which of detector 1～n is "true", the output result of other detectors is "false", According to output result be then " very " detector output result detect user sleep state, by this detector output feature The type of data is labeled as corresponding type；

In such scheme, after the mark of feature data types, these characteristics can be used for training default classification Device, thus the dormant accuracy of detection improving grader.

(2) if the output result of more than one detector is "true" in the output result of detector 1～n, or detector 1～ All output results of n are "false", then detect the sleep state of user according to the output result of default grader, and not to inspection The characteristic surveying device output is labeled；

In such scheme, because detector cannot detect, therefore can determine according to the testing result of default grader and work as The sleep state of front user, now the output characteristic data of detector 1～n cannot be used for improving the training sample of default grader This, therefore abandoned.

Step s104, the characteristic using mark is trained obtaining new grader to described default grader, and Replace described default grader, the sleep state of detection user using this new grader.

In this step, the characteristic based on mark in abovementioned steps s103, is input to default as sample It is trained in grader obtaining new grader, replace default grader with this new grader, pre- such that it is able to improve If the detection sleep state accuracy of grader.

In actual applications, continuing on with user, can persistently trigger, and constantly update grader, such that it is able to Constantly accuracy, and when being applied to other users it is also possible to re -training goes out grader, obtain being more suitable for dividing of this user Class device.

In one embodiment, when training new grader, first determine whether the quantity of the characteristic of marking types, When quantity reaches given threshold, the characteristic of mark is trained obtaining as the default grader of sample data input New grader；

By with given threshold, when the characteristic of the marking types collected reaches some, default point of input Class device is trained, it is to avoid sample size is too low, and training effect is not good.

With reference to shown in Fig. 3, Fig. 3 is that multiple Classifiers Combination detector detects dormant schematic diagram.Remove in annotation process Outside the default grader of more balance of other sample datas, also design n detector, detector 1 is used for detecting Whether user is in " sleep state 1 ", detector 2 is used for detecting in the state that whether user is in " sleep state 2 " ... ..., Detector n is used for detecting in the state that whether user is in " sleep state n ".

Default grader and detector 1～n is respectively enterd after characteristic input；Sentenced by above-mentioned inspection policies Disconnected, the current sleep state testing result of input user, for the characteristic of labeled data type, input to default grader It is trained new grader, for the characteristic of unlabeled data type, abandoned after detection.

The present invention can detect various sleep states based on the sleep state detection method of multiple Classifiers Combination, for example, When being used for the state detecting non-dynamic sleep of being sharp-eyed, the drowsy state can be detected, shallow sleep the phase, medium sleep period and deep sleep's phase etc. four Individual state, corresponds respectively to detector 1, detector 2, detector 3 and detector 4.

With reference to shown in Fig. 4, Fig. 4 is being shown based on the sleep state detecting system structure of multiple Classifiers Combination of an embodiment It is intended to, comprising:

Characteristic extraction module 101, for gathering the EEG signals that user produces in sleep procedure, according to sleep shape The identification mission of state identification extracts corresponding characteristic from described EEG signals；

Multiple Classifier Fusion detection module 102, for inputting default grader, detector 1～n respectively by described characteristic Detected；Wherein, 1～n kind sleep state of described default detection of classifier user, use respectively by corresponding detection for detector 1～n A kind of specific sleep state at family, n >=2；

Result judges data labeling module 103, if the output result for only one of which detector is true, with this inspection The output result surveying device detects the sleep state of user, and carries out type mark to this feature data；If there being multiple detectors Output result is true, or the output result of whole detector is vacation, then with the output result detection user of default grader Sleep state；

Classifier training and update module 103, for being instructed to described default grader using the characteristic of mark Get new grader, and replace described default grader, the sleep state of detection user using this new grader.

The sleep state detecting system based on multiple Classifiers Combination of the present invention and the present invention are based on multiple Classifiers Combination Sleep state detection method correspond, explain in the embodiment of the above-mentioned sleep state detection method based on multiple Classifiers Combination The technical characteristic stated and its advantage all be applied to the embodiment of the sleep state detecting system based on multiple Classifiers Combination, Hereby give notice that.

Each technical characteristic of embodiment described above can arbitrarily be combined, for making description succinct, not to above-mentioned reality The all possible combination of each technical characteristic applied in example is all described, as long as however, the combination of these technical characteristics is not deposited In contradiction, all it is considered to be the scope of this specification record.

Embodiment described above only have expressed the several embodiments of the present invention, and its description is more concrete and detailed, but simultaneously Can not therefore be construed as limiting the scope of the patent.It should be pointed out that coming for those of ordinary skill in the art Say, without departing from the inventive concept of the premise, some deformation can also be made and improve, these broadly fall into the protection of the present invention Scope.Therefore, the protection domain of patent of the present invention should be defined by claims.

Claims (10)

1. a kind of sleep state detection method based on multiple Classifiers Combination is it is characterised in that include:

The EEG signals that collection user produces in sleep procedure, the identification mission according to sleep state identification is from described brain telecommunications Corresponding characteristic is extracted in number；

Described characteristic is inputted default grader respectively, detector 1～n is detected；Wherein, described default grader inspection Survey 1～n kind sleep state of user, a kind of detector 1～n corresponding specific sleep state detecting user respectively, n >=2；

The sleep state of user if the output result of only one of which detector is true, is detected with the output result of this detector, And type mark is carried out to this feature data；If the output result having multiple detectors is true, or the output of whole detector Result is vacation, then detect the sleep state of user with the output result of default grader；

Characteristic using mark is trained obtaining new grader to described default grader, and utilizes this new classification Device replaces described default grader, the sleep state of detection user.

2. the sleep state identification model training method based on EEG signals according to claim 1 is it is characterised in that institute State default grader for detecting any one whether user be in 1～n kind sleep state, output result is " sleep state 1 " " sleep state 2 " ... or " sleep state n "；

Described detector 1 is used for detecting that in the state that whether user is in " sleep state 1 ", output result is " very ", otherwise then Output result is " vacation "；

Described detector 2 is used for detecting that in the state that whether user is in " sleep state 2 ", output result is " very ", otherwise then Output result is " vacation "；

……

Described detector n is used for detecting that in the state that whether user is in " sleep state n ", output result is " very ", otherwise then Output result is " vacation ".

If the sleep state identification model training method based on EEG signals according to claim 1 it is characterised in that The output result only one of which of detector 1～n is "true", and the output result of other detectors is "false", then according to output knot Fruit be " very " and detector output result detect user sleep state, by this detector export characteristic type mark Note as corresponding type.

If the sleep state identification model training method based on EEG signals according to claim 1 it is characterised in that In the output result of detector 1～n, the output result of more than one detector is "true", or all output knots of detector 1～n Fruit is "false", then detect the sleep state of user according to the output result of default grader, and the not spy to detector output Levy data to be labeled.

5. the sleep state detection method based on multiple Classifiers Combination according to claim 1 is it is characterised in that utilize mark The step that the characteristic of note is trained obtaining new grader to described default grader includes:

Judge the quantity of the characteristic of marking types, when quantity reaches given threshold, the characteristic of mark is made It is trained obtaining new grader for the default grader of sample data input.

6. the sleep state detection method based on multiple Classifiers Combination according to claim 1 it is characterised in that described from The step extracting corresponding characteristic in described EEG signals includes:

Extract the baseline of EEG signals, calculate the amplitude of variation of described baseline；Wherein, described amplitude of variation subtracts for baseline maximum Go minima；

After removing baseline, described EEG signals are carried out with wavelet decomposition, obtain wavelet coefficient, and calculated according to wavelet coefficient little The characteristic parameter of wave system number；Wherein, described characteristic parameter includes average, variance, kurtosis coefficient and/or the gradient system of wavelet coefficient Number；

After removing baseline, calculate lz complexity and the Sample Entropy of EEG signals；

The amplitude of variation of described baseline, the characteristic parameter of wavelet coefficient, lz complexity and Sample Entropy are set to described characteristic.

7. the sleep state detection method based on multiple Classifiers Combination according to claim 6 is it is characterised in that also wrap Include:

Described EEG signals δ wave frequency section, the signal of θ wave frequency section, α wave frequency section and β wave frequency section is extracted in wavelet reconstruction；

Calculate δ wave frequency section in EEG signals, θ wave frequency section, α wave frequency section, the energy of the β wave frequency section ratio in gross energy respectively；

Calculate respectively within the time of a frame, δ wave frequency section in EEG signals, θ wave frequency section, α wave frequency section, β ripple band energy is maximum Time span；

Described ratio and time are set to described characteristic.

8. the sleep state detection method based on multiple Classifiers Combination according to claim 7 is it is characterised in that described point Not Ji Suan δ wave frequency section in EEG signals, θ wave frequency section, α wave frequency section, the step bag of the energy of the β wave frequency section ratio in gross energy Include equation below:

r_δ=∑ (y_δ)²/p_total

r_θ=∑ (y_θ)²/p_total

r_α=∑ (y_α)²/p_total

r_β=∑ (y_β)²/p_total

Wherein p_total=∑ (y_δ)²+∑(y_θ)²+∑(y_α)²+σ(y_β)², y_δ, y_θ, y_αAnd y_βRepresent the δ frequency range after reconstruct, θ respectively The signal of frequency range, α frequency range and β frequency range, r_δ, r_θ, r_αAnd r_βRepresent δ frequency range, the energy of the signal of θ frequency range, α frequency range and β frequency range respectively Amount is in the ratio of gross energy.

9. the sleep state detection method based on multiple Classifiers Combination according to claim 7 is it is characterised in that described point Do not calculate within the time of a frame, δ wave frequency section in EEG signals, θ wave frequency section, α wave frequency section, the β ripple band energy maximum time The step of length includes equation below:

$c_{\mathrm{\δ}}=\underset{i=1}{\overset{30}{\mathrm{\σ}}}{f}_{\mathrm{\δ}}^i,f_{\mathrm{\δ}}^i=\left\{\begin{array}{c}1,if{r}_{\mathrm{\δ}}^i=\max \left(r_{\mathrm{\δ}}^i,r_{\mathrm{\θ}}^i,r_{\mathrm{\α}}^i,r_{\mathrm{\β}}^i\right)\\ 0,if{r}_{\mathrm{\δ}}^i\&notequal;\max \left(r_{\mathrm{\δ}}^i,r_{\mathrm{\θ}}^i,r_{\mathrm{\α}}^i,r_{\mathrm{\β}}^i\right)\end{array}\right.$

$c_{\mathrm{\θ}}=\underset{i=1}{\overset{30}{\mathrm{\σ}}}{f}_{\mathrm{\θ}}^i,f_{\mathrm{\θ}}^i=\left\{\begin{array}{c}1,if{r}_{\mathrm{\θ}}^i=\max \left(r_{\mathrm{\δ}}^i,r_{\mathrm{\θ}}^i,r_{\mathrm{\α}}^i,r_{\mathrm{\β}}^i\right)\\ 0,if{r}_{\mathrm{\θ}}^i\&notequal;\max \left(r_{\mathrm{\δ}}^i,r_{\mathrm{\θ}}^i,r_{\mathrm{\α}}^i,r_{\mathrm{\β}}^i\right)\end{array}\right.$

$c_{\mathrm{\α}}=\underset{i=1}{\overset{30}{\mathrm{\σ}}}{f}_{\mathrm{\α}}^i,f_{\mathrm{\α}}^i=\left\{\begin{array}{c}1,if{r}_{\mathrm{\α}}^i=\max \left(r_{\mathrm{\δ}}^i,r_{\mathrm{\θ}}^i,r_{\mathrm{\α}}^i,r_{\mathrm{\β}}^i\right)\\ 0,if{r}_{\mathrm{\α}}^i\&notequal;\max \left(r_{\mathrm{\δ}}^i,r_{\mathrm{\θ}}^i,r_{\mathrm{\α}}^i,r_{\mathrm{\β}}^i\right)\end{array}\right.$

$c_{\mathrm{\β}}=\underset{i=1}{\overset{30}{\mathrm{\σ}}}{f}_{\mathrm{\β}}^i,f_{\mathrm{\β}}^i=\left\{\begin{array}{c}1,if{r}_{\mathrm{\β}}^i=\max \left(r_{\mathrm{\δ}}^i,r_{\mathrm{\θ}}^i,r_{\mathrm{\α}}^i,r_{\mathrm{\β}}^i\right)\\ 0,if{r}_{\mathrm{\β}}^i\&notequal;\max \left(r_{\mathrm{\δ}}^i,r_{\mathrm{\θ}}^i,r_{\mathrm{\α}}^i,r_{\mathrm{\β}}^i\right)\end{array}\right.$

In formula, c_δ, c_θ, c_αAnd c_βRepresent δ frequency range, θ frequency range, α frequency range and β frequency range signal in energy proportion shared by current frame in Big time span,Represent that in i-th second, δ frequency range, the energy of the signal of θ frequency range, α frequency range and β frequency range are total respectively The ratio of energy.

10. a kind of sleep state detecting system based on multiple Classifiers Combination is it is characterised in that include:

Characteristic extraction module, for gathering the EEG signals that user produces in sleep procedure, identifies according to sleep state Identification mission extract corresponding characteristic from described EEG signals；

Multiple Classifier Fusion detection module, for described characteristic is inputted default grader respectively, detector 1～n is examined Survey；Wherein, 1～n kind sleep state of described default detection of classifier user, detector 1～n is corresponding respectively to detect the one of user Plant specific sleep state, n >=2；

Result judges data labeling module, if the output result for only one of which detector is true, with this detector Output result detects the sleep state of user, and carries out type mark to this feature data；If the output having multiple detectors is tied Fruit is true, or the output result of whole detector is vacation, then detect the sleep of user with the output result of default grader State；

Classifier training and update module, for being trained to described default grader obtaining newly using the characteristic of mark Grader, and replace described default grader, the sleep state of detection user using this new grader.