CN104665849A - Multi-physiological signal multi-model interaction-based high-speed railway dispatcher stress detecting method - Google Patents
Multi-physiological signal multi-model interaction-based high-speed railway dispatcher stress detecting method Download PDFInfo
- Publication number
- CN104665849A CN104665849A CN201510075304.9A CN201510075304A CN104665849A CN 104665849 A CN104665849 A CN 104665849A CN 201510075304 A CN201510075304 A CN 201510075304A CN 104665849 A CN104665849 A CN 104665849A
- Authority
- CN
- China
- Prior art keywords
- model
- stress state
- sample
- eye
- kinds
- Prior art date
- Legal status (The legal status 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 status listed.)
- Granted
Links
Classifications
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/24—Detecting, measuring or recording bioelectric or biomagnetic signals of the body or parts thereof
- A61B5/316—Modalities, i.e. specific diagnostic methods
- A61B5/369—Electroencephalography [EEG]
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B3/00—Apparatus for testing the eyes; Instruments for examining the eyes
- A61B3/10—Objective types, i.e. instruments for examining the eyes independent of the patients' perceptions or reactions
- A61B3/113—Objective types, i.e. instruments for examining the eyes independent of the patients' perceptions or reactions for determining or recording eye movement
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/24—Detecting, measuring or recording bioelectric or biomagnetic signals of the body or parts thereof
- A61B5/316—Modalities, i.e. specific diagnostic methods
- A61B5/318—Heart-related electrical modalities, e.g. electrocardiography [ECG]
Abstract
The invention relates to a multi-physiological signal multi-model interaction-based high-speed railway dispatcher stress detecting method. An electroencephalogram signal collecting device, a cardiograph collecting device and an eye movement collecting device are adopted to acquire corresponding physiological signals in a linkage way to be subjected to information collection, so that the stress states of high-speed railway dispatchers are judged; after characteristic selection, model training and information fusion processing, electroencephalogram, cardiograph and eye movement identification results which are analyzed on the basis of three models, i.e. a BP (Back Propagation) neural network model, a SVM (Support Vector Machine) model and an HMM (Hidden Markov Model) markov model, are subjected to information fusion, and better electroencephalogram, cardiograph and eye movement identification results are extracted; the electroencephalogram, cardiograph and eye movement identification results are fused again, and finally a fusion characteristic identification result is obtained to quickly judge the stress states of the high-speed railway dispatchers.
Description
Technical field
The present invention relates to based on brain electricity, electrocardio and eye dynamic stress detection method, particularly relating to a kind of high ferro dispatcher mutual based on many physiological signals multi-model stress detection method.
Background technology
For many years, transform through the construction to the new line of high-speed railway and the high speed to existing railway, China has had the High-speed Railway Network of maximum-norm and the highest overall trip speed in the world.The bullet train speed of service is high, density is large, once have an accident, its result will be catastrophic, and what therefore must ensure that high-speed railway runs is perfectly safe.This requires that dispatcher can grasp train operating conditions and various travelling facility situation in real time, all kinds of information jeopardizing traffic safety of timely reception, and make correct judgement and decision-making, thus effectively dispose various abnormal conditions, ensure train operating safety and normal transport order.But because people often makes false judgment and decision-making under stress state, so-called stress refer to unexpected urgent with dangerous situation under the high-strung emotional state that causes, may potential safety hazard be there is in the scheduling that therefore dispatcher makes at stress state, in view of the situation, the present invention relates to a kind of high ferro dispatcher mutual based on many physiological signals multi-model stress detection method, pass through the method, effectively can detect whether high ferro dispatcher works under stress state, thus effectively avoid dispatcher to make false judgment and decision-making under stress state.
Mostly current research is to concentrate on a kind of physiological signal, but due to the complexity of task in practical work, any single physical signs is all unilateral to the measurement of stress, only has the integrated use of multiple physical signs accurately could detect stress.Because EEG signals can more directly, more objectively react large idiophrenic activity, there is good temporal resolution, and have cannot manual control, the advantage that cannot forge, electrocardiosignal is from the depolarization and the process of repolarization that macroscopically record heart cell, objectively respond the physiological situation at each position of heart to a certain extent, the dynamic signal of eye and mental activity have direct or indirect relation, and three kinds of signals have significant correlation, therefore the present invention is based on brain electricity, electrocardio, eye move three kinds of signals has high accuracy and reliability alternately.
For brain electricity and the detection of electrocardiosignal, existing patent relates to a kind of method for detecting fatigue driving CN201410366036.1 merging brain electricity and electrocardiosignal, for detecting fatigue of automobile driver driving situation.The present invention detects dispatcher's stress state, has used BP neutral net, SVM support vector machine, HMM markov three kinds of models and information fusion technology, has had the advantage of high efficiency and high discrimination.
Summary of the invention
The object of the invention is, for the deficiency that China high ferro dispatcher stress state detects, inventing a kind of high ferro dispatcher mutual based on many physiological signals multi-model stress detection method.By high ferro dispatcher real-time brain electric data collecting device collect under dispatcher stress state with the eeg data under non-stress state, with the eye movement data under non-stress state under eye tracker collection scheduling person stress state, with the electrocardiogram (ECG) data under non-stress state under ECG collection device collection dispatcher stress state, carry out data analysis, filter out brain electricity, electrocardio, the dynamic index of eye, and use BP neutral net respectively, SVM support vector machine, HMM markov three kinds of models are further processed characteristic index, information fusion is carried out to the index after process, show that it stress fusion feature.A kind of high ferro dispatcher mutual based on many physiological signals multi-model stress detection method, adopt EEG signals gathering-device, ECG collection device, eye moves the corresponding physiological signal of harvester interlock acquisition and carries out the laggard row relax of information gathering to draw judgement to the stress state of tested high ferro dispatcher.
The object of the invention is by following means realize.
A kind of high ferro dispatcher mutual based on many physiological signals multi-model stress detection method, adopt EEG signals gathering-device, ECG collection device, the dynamic harvester interlock of eye obtains corresponding physiological signal and carries out the laggard row relax of information gathering to draw judgement to the stress state of tested high ferro dispatcher, comprise following key step
1) feature selection
Brain electricity crosslinking electrode adopts 32 to lead eeg collection system, and each electrode gathers 3 kinds of basic ripples, always has 96 characteristic indexs, obtains the strongest front 10 characteristic indexs of classification capacity from these 96 characteristic index K-W methods of inspection; Move a large amount of characteristic index with K-W method of inspection obtain the strongest front 10 characteristic indexs of classification capacity from electrocardio, eye respectively.
Wherein as follows for brain electrical feature choose targets:
For an arbitrary parameter x
j, it is mixed with the argument sequence obtained under non-stress state at stress state, carries out Kruskal-Wallis inspection:
In formula, H is statistic of test, and i is stress state grade scalar i=1 is non-stress state, and i=2 is stress state;
under representing the i-th class stress state grade, the average order of parameter sample; M is the total amount of two class parameter samples;
Because H obeys card side's distribution that degree of freedom is 1, table look-up and can obtain the Probability p of critical region, p value represents the identical probability of sample distribution under two kinds of stress state conditions, therefore p value is less, and to represent the DATA DISTRIBUTION difference of this parameter under two class stress state larger, based on said method, electroencephalogram parameter is tested one by one, from multinomial electroencephalogram parameter, selects the most significant q item of difference form driving Mental Workload characteristic index, and constitutive characteristic vector X={x
1, x
2, x
3... x
q, };
2) model training
Randomly draw the sample of 75% as training sample, the sample of 25% is as test sample book; Respectively to three kinds of models, that is: BP neural network model, SVM supporting vector machine model and HHM Markov model give training and testing;
Suitable BP neural network model, SVM supporting vector machine model, HHM Markov model is found respectively by model training.Respectively brain electricity, electrocardio, eye movement characteristics index are better classified according to the BP neural network model trained, SVM supporting vector machine model, HHM Markov model, export two kinds of results, and specifying that non-stress state exports is 0, and it is 1 that stress state exports.
3) information fusion
Respectively to 2) obtain move recognition result carry out information fusion based on BP neural network model, SVM supporting vector machine model, the brain electricity of HHM Markov model three kinds of model analysiss, electrocardio, eye, extract better brain electricity, electrocardio, eye move recognition result, again brain electricity, electrocardio, eye are moved recognition result and merged, finally obtain comprehensive fusion feature value; High ferro dispatcher stress state can be judged fast and effectively by final fusion feature recognition result; Information fusion adopts weighted mean method, and information be weighted on average, result is worth as fusion.
In the described information fusion stage, also can adopt pre-set threshold value, comprehensive fusion feature value was less than the non-stress state of threshold values at that time, was greater than threshold values and judged stress state.
Compared with prior art, the present invention considers brain electricity, electrocardio, the dynamic three kinds of signals of eye and BP neural network model, with SVM supporting vector machine model, HMM Markov model three kinds of models, BP neutral net have extremely strong non-linear mapping capability and to external world input amendment have very strong recognition and classification ability, SVM supporting vector machine model is at solution small sample, non-linear and high dimensional pattern identification problem has distinctive advantage, HMM has very successful application at area of pattern recognition, what finally draw based on these three kinds of models characteristic index stress have higher accuracy and reliability.
The technical solution adopted in the present invention comprises information acquisition module, feature selection module, model training module, information fusion module.
Accompanying drawing explanation
The mutual high ferro dispatcher of Fig. 1 many physiological signals multi-model stress detection module figure
The mutual high ferro dispatcher of Fig. 2 many physiological signals multi-model stress overhaul flow chart
Fig. 3 eeg signal acquisition installation drawing
Fig. 4 electrocardiogram acquisition schematic diagram
Detailed description of the invention
Below in conjunction with drawings and Examples, the present invention is described in further detail.It should be noted that, specific embodiment described herein is only for explaining the present invention, but not the restriction to application scenarios of the present invention.In addition, for convenience of description, part related to the present invention is only illustrated but not full content in accompanying drawing.
Realize the hardware foundation of information gathering of the present invention: information acquisition module
This module comprises high ferro dispatcher EEG signals gathering-device, ECG collection device, and eye moves harvester.
High ferro dispatcher EEG signals gathering-device comprises the electrode for encephalograms that leads, signal amplifier, wave filter, processor and EEG signals storage device.
EEG signals when high ferro dispatcher EEG signals gathering-device real-time collecting high ferro dispatcher works, and EEG signals is outputted to amplifier, the EEG signals of amplification for amplifying EEG signals, and is outputted to wave filter by amplifier; Signal after filtering for filtering out useless signal, and is outputted to processor by wave filter.
ECG collection device comprises ECG collection device electrode, heart rate sensor, signal amplifier, wave filter, processor.
The dynamic harvester of eye is eye tracker, collects the eye movement data of dispatcher.
Like this, require mental skill electricity, electrocardio, ophthalmogyric device gathers brain electricity, electrocardio, the eye movement data of high ferro dispatcher under stress state and non-stress state respectively.Be below the concrete grammar of date processing:
1. feature selection
The object of feature selection from a large amount of features, selects the effective feature of part or forms minority new feature, with description scheme structure effectively or the separability improving all kinds of sample, thus uses manpower and material resources sparingly, to obtain better economic benefit.
The present invention utilizes K-W (the kruskal and walis) method of inspection from original feature, select part feature preferably, gives up the feature that effect is poor.K-W (kruskal and walis) inspection is a kind of conventional single feature selection approach.Its advantage is: 1) implement more convenient, amount of calculation is little, and cost is few; 2) it selects Partial Feature from former feature, and therefore comparatively directly perceived, physical significance is also clearer.
Brain electricity crosslinking electrode adopts 32 to lead eeg collection system, and each electrode gathers 3 kinds of basic ripples, always has 96 characteristic indexs, obtains the strongest front 10 characteristic indexs of classification capacity from these 96 characteristic index K-W methods of inspection.Move a large amount of characteristic index with K-W method of inspection obtain the strongest front 10 characteristic indexs of classification capacity from electrocardio, eye respectively.
Wherein as follows for brain electrical feature choose targets:
For improving accuracy of identification, reducing characteristic vector dimension simultaneously, reducing data operation load amount, from multinomial electroencephalogram parameter, some parameters need be chosen as detecting characteristic index.Specific practice is, for an arbitrary parameter x
j, it is mixed with the argument sequence obtained under non-stress state at stress state, carries out Kruskal-Wallis inspection:
In formula, H is statistic of test, and i is stress state grade scalar i=1 is non-stress state, and i=2 is stress state, n
ibe i-th sample;
under representing the i-th class stress state grade, the average order of parameter sample; M is the total amount of two class parameter samples.
Because H obeys card side's distribution that degree of freedom is 1, table look-up and can obtain the Probability p of critical region.P value represents the identical probability of sample distribution under two kinds of stress state conditions, therefore p value is less, and to represent the DATA DISTRIBUTION difference of this parameter under two class stress state larger.Based on said method, electroencephalogram parameter is tested one by one, from multinomial electroencephalogram parameter, select the most significant q item of difference form driving Mental Workload characteristic index, and constitutive characteristic vector X={x
1, x
2, x
3... x
q, }.
2. model training
Randomly draw the sample of 75% as training sample, the sample of 25% is as test sample book.Training and testing is given to three kinds of models.
The wherein connection weights being trained for neutral net continuous change network under the stimulation of extraneous input amendment of BP neural network model, to make the output of network constantly close to the output expected.
It adopts error back propagation (BP) to adjust weight coefficient neural network training, thus total error is reduced gradually, completes network model's training.
For the training of SVM supporting vector machine model, concentrate from candidate samples randomly and select a small amount of sample and the classification marking them, structure initial training sample set, guarantee in initial training sample set, at least to include a positive example sample and a negative routine sample.Initial training sample set is utilized to train a grader.Under this grader, adopt certain sampling algorithm, concentrate selection to be conducive to the sample of classifier performance most from candidate samples, mark classification and join training sample and concentrate, re-training grader.Repeat this process, until candidate samples collection is empty or reaches certain index.Finally find suitable nonlinear mapping function.
For the training of HHM Markov model, adjustment model parameter, makes output probability maximum.
1. initial model (treating training pattern) λ
0,
2. based on λ
0and observed value sequence X, training new model λ,
3. if logP (X/ λ)-logP (X/ λ
0) > δ, illustrate that training produces a desired effect, algorithm terminates.
4. otherwise, make λ
0=λ, continues the 2nd step work.
Suitable BP neural network model, SVM supporting vector machine model, HHM Markov model is found respectively by model training.Respectively brain electricity, electrocardio, eye movement characteristics index are better classified according to the BP neural network model trained, SVM supporting vector machine model, HHM Markov model, export two kinds of results, and specifying that non-stress state exports is 0, and it is 1 that stress state exports.
3. information fusion
Information fusion carries out multi-level, many-sided, multi-level process to multiple sensing data, namely combines or merge the data from multiple sensor or other information sources, comprehensive to obtain, and better estimates.
Move recognition result carry out information fusion to based on BP neural network model, SVM supporting vector machine model, the brain electricity of HHM Markov model three kinds of model analysiss, electrocardio, eye respectively, extract better brain electricity, electrocardio, eye move recognition result, again brain electricity, electrocardio, eye are moved recognition result and merged, finally obtain fusion feature recognition result.High ferro dispatcher stress state can be judged fast and effectively by final fusion feature recognition result.
This information fusion adopts weighted mean method, and information be weighted on average, result is worth as fusion.As used for eeg data three kinds of model identifying devices by generation three kinds of recognition results respectively, giving different weighted values to three kinds of models, three kinds of recognition results are weighted on average, obtain the recognition result after merging.Move and give different weights to again brain electricity, electrocardio, eye, calculate the fusion recognition result that it is total, judge the stress state of high ferro dispatcher according to final fusion fusion feature recognition result.
Embodiment one
As shown in Figure 1, 2, this example provides a kind of high ferro dispatcher mutual based on many physiological signals multi-model stress detection method.
The technical solution adopted in the present invention comprises information acquisition module 100, feature selection module 200, model training module 300, information fusion module 400.
Step one
Information acquisition module comprises the real-time EEG signals gathering-device 101 of high ferro dispatcher, electrocardiosignal gathering-device 102, eye movement data harvester 103.
1. eeg signal acquisition device 101 comprises as shown in Figure 3: the electrode for encephalograms that leads, amplifier, wave filter, computer and LCDs.
The outfan of brain electrode of leading is connected with the input of described amplifier, and the outfan of described amplifier is connected with the input of described wave filter, and the outfan of described wave filter is connected with described processor, and the outfan of described processor is connected with LCDs.
The electrode for encephalograms that leads is used for connecting drives experimenter, takes high ferro dispatcher EEG signals, adopts 32 to lead eeg collection system (Brain Products, GmbH, Munich, Germany) and carry out continuous acquisition to EEG data.Recording electrode is with reference to international electroencephalology meeting 10/20 modular system (Fisch, 2000; Stern & Engel, 2005), all electrodes for physical reference, record electro-ocular signal with FCZ electrode simultaneously.Impedance between scalp and motor is less than 5k Ω, and record bandwidth is 0.5-100Hz, and sample rate is 1000Hz, and EEG signals is outputted to amplifier;
Amplifier is used for amplifying described signal, and the signal after amplifying is outputted to described wave filter, because EEG signals is a kind of extremely faint bioelectrical signals, after acquired signal, need to amplify signal, thus the interfering signal of certain limit can be got rid of.
Wave filter is used for filtering the signal after described amplification, and the signal after filtering is outputted to described processor, and the process of brain wave acquisition inevitably will be subject to the effect of noise such as electrocardio noise, noise of equipment.So, filtering is carried out to eeg data very important.
Processor is used for the EEG signals after to filtration and carries out pretreatment, artefact removal etc.
Liquid crystal display is used for showing EEG oscillogram.
2. ECG collection device 102 comprises electrode, heart rate sensor, signal amplifier, wave filter, processor.
The input of Electrode connection heart rate sensor, the input of the outfan connection signal amplifier of heart rate sensor, the outfan of signal amplifier is electrically connected with the input of wave filter, and the outfan of wave filter is connected with processor.Wherein electrode with the connected mode of dispatcher as shown in Figure 4, adopt three to lead system, three conducting wire cable wraps are containing the right lower limb of RR right arm, LR left arm and RL three conducting wires, and LR pole of wherein leading connects left upper extremity, and RR connects right upper extremity, composition bipolar lead.When careful electrostimulation is conducted from right to left, the voltage of the voltage ratio right upper extremity electrode slice of left upper extremity electrode slice is high, can produce a forward ecg wave form.
3. a dynamic harvester 103 is eye tracker, collects the eye movement data of dispatcher.
Require mental skill electricity, electrocardio, ophthalmogyric device gathers brain electricity, electrocardio, the eye movement data of high ferro dispatcher under stress state and non-stress state respectively.
Step 2
Feature selection module 200
Feature selection refers to the process forming or select minority validity feature from the original feature extracted further.The standard selected is only just can description scheme structure and improve the separability of all kinds of sample by a few features.The mode selected is broadly divided into two classes: single feature selection and dimensionality reduction map.The former selects Partial Feature from original feature, and the latter refers to carry out certain conversion to obtain new feature to each original feature.This patent adopts the K-W method of inspection, single feature selection approach.The method algorithm is as follows:
1. pair whole sample, according to the value of feature under each sample jth, gives each sample number into spectrum by order from small to large.
Such as, X
jthe minimum sample of value can be numbered 1, and secondary little sample can be numbered 2, so continues, and last sample being numbered N is maximum that the lower sample of value.
If there is the X corresponding to n sample
jbe worth identical, to these sample random number.Such as, discharge for first 3 numbers, and the minimum sample value of current value has three, their X
jbeing worth equal, then can be No. 4 one of them sample row with rule, and a row taken out by random machine to two remaining samples is again No. 5, and a finally remaining row is No. 6.
2. the meansigma methods of every apoplexy due to endogenous wind each sample numbering, is designated as respectively
3. compute statistics m, formula is:
M meets the X that degree of freedom is K 1
2distribution.
In above formula bracket, (N+1)/2 are averages of all sample number into spectrum, and R
ithe average of the i-th class numbering, the deviation of all kinds of that what therefore m measured is.M is larger, and between representation class, deviation difference is larger, namely all kinds ofly can separate preferably.Therefore, we think the feature x that m is larger
j, its classification capacity is stronger.
With K-W (kruskal and walis) inspection, single feature selection is carried out, the therefrom relatively good feature of selective power to the feature that brain electricity, electrocardio, eye move extraction.
Step 3
Model training 300
Randomly draw the sample of 75% as training sample, the sample of 25% is as test sample book.Training and testing is given to three kinds of models.
Wherein BP trains principle
Utilize the error after exporting estimate output layer directly before the error of conducting shell, then by the error of the more front one deck of this error estimation, anti-pass so is in layer gone down, and just obtains the error estimation of every other each layer.
The first step, netinit
The random number in an interval (-1,1) is composed respectively, specification error function e, given computational accuracy value ε and maximum study number of times M to each connection weights.
Second step, a random selecting kth input amendment and corresponding desired output.
x(k)＝(x
1(k),x
2(k),…,x
n(k))
d
o(k)＝(d
1(k),d
2(k),…,d
q(k))
3rd step, calculates each neuronic input and output of hidden layer.
ho
h(k)＝f(hi
h(k)) h＝1,2,…,p
yo
o(k)＝f(yi
o(k)) o＝1,2,…q
4th step, utilize network desired output and actual output, error of calculation function is to each neuronic partial derivative δ of output layer
o(k).
5th step, utilizes the connection weights of hidden layer to output layer, the δ of output layer
ok the output error of calculation function of () and hidden layer is to each neuronic partial derivative δ of hidden layer
h(k).
6th step, utilizes each neuronic δ of output layer
ok connection weight w is revised in () and each neuronic output of hidden layer
ho(k).
7th step, utilizes each neuronic δ of hidden layer
h(k) and each neuronic Introduced Malaria connection weight of input layer.
8th step, calculates global error.
9th step, judges whether network error meets the demands.When error reaches default precision or study number of times is greater than the maximum times of setting, then terminate algorithm.Otherwise, choose the desired output of next learning sample and correspondence, turn back to the 3rd step, enter next round study.
Variable-definition: input vector; X=(x
1, x
2..., x
n) hidden layer input vector; X=(x
1, x
2..., x
n)
Hidden layer output vector; Ho=(ho
1, ho
2..., ho
p) output layer input vector; Yi=(yi
1, yi
2..., yi
q)
Output layer output vector; Yo=(yo
1, yo
2..., yo
q) desired output vector; d
o=(d
1, d
2..., d
q)
The connection weights in input layer and intermediate layer: w
ihthe connection weights of hidden layer and output layer: w
ho
The each neuronic threshold value of hidden layer: b
hthe each neuronic threshold value of output layer: b
o
Sample data number: k=1,2 ... m activation primitive: f ()
Error function:
The arthmetic statement that the SVM of Active Learning is concrete is as follows:
1., according to initial condition structure initial training sample set, ensure at least to include a positive example sample and a negative routine sample.
2. find optimal separating hyper plane according to known training sample set, design SVM classifier.
3. if still have sample point in contiguous with separating surface gap, then select the sample from classification boundaries is nearest to evaluate, this sample is added training sample set, and get back to the 2nd step.
4. from whole training sample, repeat Stochastic choice sample evaluate, and this sample is added training set (evaluated sample is only counted, need not again evaluate), grader is utilized to evaluate sample, if the evaluation result of grader and authentic assessment inconsistent, then get back to the 2nd step.
5. repeat the 4th step, if N continuous time is evaluated consistent, algorithm stops.
Suitable BP neural network model, SVM supporting vector machine model, HHM Markov model is found respectively by model training.Respectively brain electricity, electrocardio, eye movement characteristics index are better classified according to the BP neural network model trained, SVM supporting vector machine model, HHM Markov model.
For the training of HHM Markov model, adjustment model parameter, makes output probability maximum.
1. initial model (treating training pattern) λ
0,
2. based on λ
0and observed value sequence X, training new model λ,
3. if logP (X/ λ)-logP (X/ λ
0) > δ, illustrate that training produces a desired effect, algorithm
Terminate.
4. otherwise, make λ
0=λ, continues the 2nd step work.
Suitable BP neural network model, SVM supporting vector machine model, HHM Markov model is found respectively by model training.Respectively brain electricity, electrocardio, eye movement characteristics index are better classified according to the BP neural network model trained, SVM supporting vector machine model, HHM Markov model, export two kinds of results, and specifying that non-stress state exports is 0, and it is 1 that stress state exports.
Two kinds of recognition results stress be exported by the model that trains of recognition feature vector sample input.
Step 4
Information fusion 400
Information fusion carries out multi-level, many-sided, multi-level process to multiple sensing data, namely combines or merge the data from multiple sensor or other information sources, comprehensive to obtain, and better estimates.
Move recognition result carry out information fusion to based on BP neural network model, SVM supporting vector machine model, the brain electricity of HHM Markov model three kinds of model analysiss, electrocardio, eye respectively, extract better brain electricity, electrocardio, eye move recognition result, again brain electricity, electrocardio, eye are moved recognition result and merged, finally obtain fusion feature recognition result.High ferro dispatcher stress state can be judged fast and effectively by final fusion feature recognition result.
This information fusion adopts weighted mean method, and information be weighted on average, result is worth as fusion.As used for eeg data three kinds of model identifying devices by generation three kinds of recognition results respectively, giving different weighted values to three kinds of models, three kinds of recognition results are weighted on average, obtain the recognition result after merging.Move and give different weights to again brain electricity, electrocardio, eye, calculate the fusion recognition result that it is total, judge the stress state of high ferro dispatcher according to final fusion fusion feature recognition result.
Claims (2)
1. one kind stress detection method based on the mutual high ferro dispatcher of many physiological signals multi-model, adopt EEG signals gathering-device, ECG collection device, the dynamic harvester interlock of eye obtains corresponding physiological signal and carries out the laggard row relax of information gathering to draw judgement to the stress state of tested high ferro dispatcher, comprise following key step
1) feature selection
Brain electricity crosslinking electrode adopts 32 to lead eeg collection system, and each electrode gathers 3 kinds of basic ripples, always has 96 characteristic indexs, obtains the strongest front 10 characteristic indexs of classification capacity from these 96 characteristic index K-W methods of inspection; Move a large amount of characteristic index with K-W method of inspection obtain the strongest front 10 characteristic indexs of classification capacity from electrocardio, eye respectively;
Wherein as follows for brain electrical feature choose targets:
For an arbitrary parameter x
j, it is mixed with the argument sequence obtained under non-stress state at stress state, carries out Kruskal-Wallis inspection:
In formula, H is statistic of test, and i is stress state grade scalar i=1 is non-stress state, and i=2 is stress state;
under representing the i-th class stress state grade, the average order of parameter sample; M is the total amount of two class parameter samples;
Because H obeys card side's distribution that degree of freedom is 1, table look-up and can obtain the Probability p of critical region, p value represents the identical probability of sample distribution under two kinds of stress state conditions, therefore p value is less, and to represent the DATA DISTRIBUTION difference of this parameter under two class stress state larger, based on said method, electroencephalogram parameter is tested one by one, from multinomial electroencephalogram parameter, selects the most significant q item of difference form driving Mental Workload characteristic index, and constitutive characteristic vector x={ x
1, x
2, x
3... x
q, };
2) model training
Randomly draw the sample of 75% as training sample, the sample of 25% is as test sample book; Respectively to three kinds of models, that is: BP neural network model, SVM supporting vector machine model and HHM Markov model give training and testing;
Suitable BP neural network model, SVM supporting vector machine model, HHM Markov model is found respectively by model training.Respectively brain electricity, electrocardio, eye movement characteristics index are better classified according to the BP neural network model trained, SVM supporting vector machine model, HHM Markov model, export two kinds of results, and specifying that non-stress state exports is 0, and it is 1 that stress state exports;
3) information fusion
Respectively to 2) obtain move recognition result carry out information fusion based on BP neural network model, SVM supporting vector machine model, the brain electricity of HHM Markov model three kinds of model analysiss, electrocardio, eye, extract better brain electricity, electrocardio, eye move recognition result, again brain electricity, electrocardio, eye are moved recognition result and merged, finally obtain comprehensive fusion feature value; High ferro dispatcher stress state can be judged fast and effectively by final fusion feature recognition result; Information fusion adopts weighted mean method, and information be weighted on average, result is worth as fusion.
2. the high ferro dispatcher mutual based on many physiological signals multi-model according to claim 1 stress detection method, it is characterized in that, the described information fusion stage, adopt pre-set threshold value, comprehensive fusion feature value was less than the non-stress state of threshold values at that time, was greater than threshold values and judged stress state.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510075304.9A CN104665849B (en) | 2014-12-11 | 2015-02-12 | A kind of high ferro dispatcher based on the interaction of many physiological signals multi-model stress detection method |
Applications Claiming Priority (3)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410763496 | 2014-12-11 | ||
| CN2014107634968 | 2014-12-11 | ||
| CN201510075304.9A CN104665849B (en) | 2014-12-11 | 2015-02-12 | A kind of high ferro dispatcher based on the interaction of many physiological signals multi-model stress detection method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN104665849A true CN104665849A (en) | 2015-06-03 |
| CN104665849B CN104665849B (en) | 2017-03-01 |
Family
ID=53301822
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510075304.9A Active CN104665849B (en) | 2014-12-11 | 2015-02-12 | A kind of high ferro dispatcher based on the interaction of many physiological signals multi-model stress detection method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104665849B (en) |
Cited By (27)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105534517A (en) * | 2016-02-29 | 2016-05-04 | 浙江铭众科技有限公司 | Method for removing vehicle motion noise in three-lead electrocardiosignal |
| CN105769173A (en) * | 2016-02-29 | 2016-07-20 | 浙江铭众科技有限公司 | Electrocardiogram monitoring system with electrocardiosignal denoising function |
| CN105796091A (en) * | 2016-02-29 | 2016-07-27 | 浙江铭众科技有限公司 | Intelligent terminal for removing electrocardiosignal vehicle motion noise |
| CN105877766A (en) * | 2016-06-21 | 2016-08-24 | 东北大学 | Mental state detection system and method based on multiple physiological signal fusion |
| CN107220669A (en) * | 2017-05-27 | 2017-09-29 | 西南交通大学 | The method of testing and system of the behavior monitoring ability of dispatcher a kind of |
| CN107220772A (en) * | 2017-05-27 | 2017-09-29 | 西南交通大学 | The method of testing and system of high ferro dispatch's task's switching capability |
| CN107252313A (en) * | 2017-05-25 | 2017-10-17 | 深圳市卡迪赛克科技有限公司 | The monitoring method and system of a kind of safe driving, automobile, readable storage medium storing program for executing |
| CN107536617A (en) * | 2016-06-27 | 2018-01-05 | 现代自动车株式会社 | The apparatus and method of bio-identification signal detection driver status based on driver |
| CN108038456A (en) * | 2017-12-19 | 2018-05-15 | 中科视拓（北京）科技有限公司 | A kind of anti-fraud method in face identification system |
| CN108241431A (en) * | 2016-12-26 | 2018-07-03 | 中国移动通信有限公司研究院 | A kind of task adjusting method and device |
| CN108281184A (en) * | 2018-02-01 | 2018-07-13 | 中国科学院上海微系统与信息技术研究所 | Myocardial ischemia based on machine learning examines method, storage medium and electronic equipment in advance |
| CN109145890A (en) * | 2018-11-19 | 2019-01-04 | 东南大学 | A kind of model of the driver psychology stress intensity intelligent recognition based on single mode physiological signal |
| CN109255309A (en) * | 2018-08-28 | 2019-01-22 | 中国人民解放军战略支援部队信息工程大学 | Brain electricity and eye movement fusion method and device towards Remote Sensing Target detection |
| CN109620265A (en) * | 2018-12-26 | 2019-04-16 | 中国科学院深圳先进技术研究院 | Recognition methods and relevant apparatus |
| CN109976525A (en) * | 2019-03-27 | 2019-07-05 | 上海大学 | A kind of user interface interaction method, apparatus and computer equipment |
| CN110123266A (en) * | 2019-05-05 | 2019-08-16 | 北京航空航天大学 | A kind of maneuvering decision modeling method based on multi-modal physiologic information |
| CN110507319A (en) * | 2019-09-04 | 2019-11-29 | 杭州回车电子科技有限公司 | A kind of fatigue strength detection method and equipment |
| CN110811595A (en) * | 2019-11-28 | 2020-02-21 | 中国银行股份有限公司 | Abnormal state detection method and device, electronic equipment and computer storage medium |
| CN110811649A (en) * | 2019-10-31 | 2020-02-21 | 太原理工大学 | Fatigue driving detection method based on bioelectricity and behavior characteristic fusion |
| CN111078984A (en) * | 2019-11-05 | 2020-04-28 | 深圳奇迹智慧网络有限公司 | Network model publishing method and device, computer equipment and storage medium |
| CN112033335A (en) * | 2020-11-05 | 2020-12-04 | 成都中轨轨道设备有限公司 | Intelligent monitoring and early warning system and method for railway gauging rule |
| CN113179970A (en) * | 2021-04-09 | 2021-07-30 | 中国农业大学 | Method and system for distinguishing stress state of fish |
| CN113349780A (en) * | 2021-06-07 | 2021-09-07 | 浙江科技学院 | Method for evaluating influence of emotional design on online learning cognitive load |
| CN113627740A (en) * | 2021-07-20 | 2021-11-09 | 东风汽车集团股份有限公司 | Driving load evaluation model construction system and construction method |
| CN113951903A (en) * | 2021-10-29 | 2022-01-21 | 西南交通大学 | High-speed railway dispatcher overload state identification method based on electroencephalogram data determination |
| CN115017996A (en) * | 2022-06-15 | 2022-09-06 | 中国铁道科学研究院集团有限公司 | Mental load prediction method and system based on multiple physiological parameters |
| CN115067944A (en) * | 2022-08-22 | 2022-09-20 | 深圳市心流科技有限公司 | Eye movement state evaluation method and device, terminal equipment and storage medium |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107095670A (en) * | 2017-05-27 | 2017-08-29 | 西南交通大学 | Time of driver's reaction Forecasting Methodology |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20030028327A1 (en) * | 2001-05-15 | 2003-02-06 | Daniela Brunner | Systems and methods for monitoring behavior informatics |
| CN1934596A (en) * | 2004-03-22 | 2007-03-21 | 沃尔沃技术公司 | Method and system for perceptual suitability test of a driver |
| CN101049236A (en) * | 2007-05-09 | 2007-10-10 | 西安电子科技大学 | Instant detection system and detection method for state of attention based on interaction between brain and computer |
| US20110201951A1 (en) * | 2010-02-12 | 2011-08-18 | Siemens Medical Solutions Usa, Inc. | System for cardiac arrhythmia detection and characterization |
| CN102697494A (en) * | 2012-06-14 | 2012-10-03 | 西南交通大学 | Wearable electroencephalogram signal collection equipment for high-speed train drivers |
| CN104127195A (en) * | 2014-07-29 | 2014-11-05 | 杭州电子科技大学 | Electroencephalogram signal and electrocardiogram signal integrated method for detecting fatigue driving |
-
2015
- 2015-02-12 CN CN201510075304.9A patent/CN104665849B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20030028327A1 (en) * | 2001-05-15 | 2003-02-06 | Daniela Brunner | Systems and methods for monitoring behavior informatics |
| CN1934596A (en) * | 2004-03-22 | 2007-03-21 | 沃尔沃技术公司 | Method and system for perceptual suitability test of a driver |
| CN101049236A (en) * | 2007-05-09 | 2007-10-10 | 西安电子科技大学 | Instant detection system and detection method for state of attention based on interaction between brain and computer |
| US20110201951A1 (en) * | 2010-02-12 | 2011-08-18 | Siemens Medical Solutions Usa, Inc. | System for cardiac arrhythmia detection and characterization |
| CN102697494A (en) * | 2012-06-14 | 2012-10-03 | 西南交通大学 | Wearable electroencephalogram signal collection equipment for high-speed train drivers |
| CN104127195A (en) * | 2014-07-29 | 2014-11-05 | 杭州电子科技大学 | Electroencephalogram signal and electrocardiogram signal integrated method for detecting fatigue driving |
Non-Patent Citations (1)
| Title |
|---|
| 刘玉增: "驾驶过程的动态仿真分析", 《交通运输工程与信息学报》 * |
Cited By (38)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105769173B (en) * | 2016-02-29 | 2018-08-28 | 浙江铭众科技有限公司 | A kind of cardioelectric monitor system with electrocardiosignal noise removal function |
| CN105769173A (en) * | 2016-02-29 | 2016-07-20 | 浙江铭众科技有限公司 | Electrocardiogram monitoring system with electrocardiosignal denoising function |
| CN105796091A (en) * | 2016-02-29 | 2016-07-27 | 浙江铭众科技有限公司 | Intelligent terminal for removing electrocardiosignal vehicle motion noise |
| CN105534517A (en) * | 2016-02-29 | 2016-05-04 | 浙江铭众科技有限公司 | Method for removing vehicle motion noise in three-lead electrocardiosignal |
| CN105796091B (en) * | 2016-02-29 | 2018-09-28 | 浙江铭众科技有限公司 | A kind of intelligent terminal for removing electrocardiosignal vehicle movement noise |
| CN105877766A (en) * | 2016-06-21 | 2016-08-24 | 东北大学 | Mental state detection system and method based on multiple physiological signal fusion |
| CN105877766B (en) * | 2016-06-21 | 2019-02-05 | 东北大学 | A kind of state of mind detection system and method based on the fusion of more physiological signals |
| CN107536617A (en) * | 2016-06-27 | 2018-01-05 | 现代自动车株式会社 | The apparatus and method of bio-identification signal detection driver status based on driver |
| CN108241431B (en) * | 2016-12-26 | 2020-10-30 | 中国移动通信有限公司研究院 | Task adjusting method and device |
| CN108241431A (en) * | 2016-12-26 | 2018-07-03 | 中国移动通信有限公司研究院 | A kind of task adjusting method and device |
| CN107252313A (en) * | 2017-05-25 | 2017-10-17 | 深圳市卡迪赛克科技有限公司 | The monitoring method and system of a kind of safe driving, automobile, readable storage medium storing program for executing |
| CN107220772A (en) * | 2017-05-27 | 2017-09-29 | 西南交通大学 | The method of testing and system of high ferro dispatch's task's switching capability |
| CN107220669A (en) * | 2017-05-27 | 2017-09-29 | 西南交通大学 | The method of testing and system of the behavior monitoring ability of dispatcher a kind of |
| CN108038456A (en) * | 2017-12-19 | 2018-05-15 | 中科视拓（北京）科技有限公司 | A kind of anti-fraud method in face identification system |
| CN108038456B (en) * | 2017-12-19 | 2024-01-26 | 中科视拓（北京）科技有限公司 | Anti-deception method in face recognition system |
| CN108281184A (en) * | 2018-02-01 | 2018-07-13 | 中国科学院上海微系统与信息技术研究所 | Myocardial ischemia based on machine learning examines method, storage medium and electronic equipment in advance |
| CN109255309A (en) * | 2018-08-28 | 2019-01-22 | 中国人民解放军战略支援部队信息工程大学 | Brain electricity and eye movement fusion method and device towards Remote Sensing Target detection |
| CN109145890A (en) * | 2018-11-19 | 2019-01-04 | 东南大学 | A kind of model of the driver psychology stress intensity intelligent recognition based on single mode physiological signal |
| CN109620265A (en) * | 2018-12-26 | 2019-04-16 | 中国科学院深圳先进技术研究院 | Recognition methods and relevant apparatus |
| CN109976525A (en) * | 2019-03-27 | 2019-07-05 | 上海大学 | A kind of user interface interaction method, apparatus and computer equipment |
| CN110123266A (en) * | 2019-05-05 | 2019-08-16 | 北京航空航天大学 | A kind of maneuvering decision modeling method based on multi-modal physiologic information |
| CN110507319A (en) * | 2019-09-04 | 2019-11-29 | 杭州回车电子科技有限公司 | A kind of fatigue strength detection method and equipment |
| CN110507319B (en) * | 2019-09-04 | 2022-04-15 | 杭州回车电子科技有限公司 | Fatigue degree detection method and equipment |
| CN110811649A (en) * | 2019-10-31 | 2020-02-21 | 太原理工大学 | Fatigue driving detection method based on bioelectricity and behavior characteristic fusion |
| CN111078984A (en) * | 2019-11-05 | 2020-04-28 | 深圳奇迹智慧网络有限公司 | Network model publishing method and device, computer equipment and storage medium |
| CN111078984B (en) * | 2019-11-05 | 2024-02-06 | 深圳奇迹智慧网络有限公司 | Network model issuing method, device, computer equipment and storage medium |
| CN110811595A (en) * | 2019-11-28 | 2020-02-21 | 中国银行股份有限公司 | Abnormal state detection method and device, electronic equipment and computer storage medium |
| CN110811595B (en) * | 2019-11-28 | 2022-04-08 | 中国银行股份有限公司 | Abnormal state detection method and device, electronic equipment and computer storage medium |
| CN112033335A (en) * | 2020-11-05 | 2020-12-04 | 成都中轨轨道设备有限公司 | Intelligent monitoring and early warning system and method for railway gauging rule |
| CN113179970B (en) * | 2021-04-09 | 2022-03-22 | 中国农业大学 | Method and system for distinguishing stress state of fish |
| CN113179970A (en) * | 2021-04-09 | 2021-07-30 | 中国农业大学 | Method and system for distinguishing stress state of fish |
| CN113349780A (en) * | 2021-06-07 | 2021-09-07 | 浙江科技学院 | Method for evaluating influence of emotional design on online learning cognitive load |
| CN113627740A (en) * | 2021-07-20 | 2021-11-09 | 东风汽车集团股份有限公司 | Driving load evaluation model construction system and construction method |
| CN113951903A (en) * | 2021-10-29 | 2022-01-21 | 西南交通大学 | High-speed railway dispatcher overload state identification method based on electroencephalogram data determination |
| CN113951903B (en) * | 2021-10-29 | 2022-07-08 | 西南交通大学 | High-speed railway dispatcher overload state identification method based on electroencephalogram data determination |
| CN115017996A (en) * | 2022-06-15 | 2022-09-06 | 中国铁道科学研究院集团有限公司 | Mental load prediction method and system based on multiple physiological parameters |
| CN115067944A (en) * | 2022-08-22 | 2022-09-20 | 深圳市心流科技有限公司 | Eye movement state evaluation method and device, terminal equipment and storage medium |
| CN115067944B (en) * | 2022-08-22 | 2022-11-11 | 深圳市心流科技有限公司 | Eye movement state evaluation method and device, terminal equipment and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN104665849B (en) | 2017-03-01 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN104665849A (en) | Multi-physiological signal multi-model interaction-based high-speed railway dispatcher stress detecting method | |
| CN102119857B (en) | Electroencephalogram detecting system and method for fatigue driving on basis of matching pursuit algorithm | |
| CN105877766B (en) | A kind of state of mind detection system and method based on the fusion of more physiological signals | |
| CN102319063B (en) | Device and method for improving acceleration identification accuracy of fetal heart rate data | |
| CN106405339A (en) | Power transmission line fault reason identification method based on high and low frequency wavelet feature association | |
| CN202404166U (en) | On-line monitoring system for vibration performance of transformer | |
| CN104887224A (en) | Epileptic feature extraction and automatic identification method based on electroencephalogram signal | |
| CN104720796A (en) | Automatic detecting system and method for epileptic attack time period | |
| CN105595990A (en) | Intelligent terminal device for evaluating and distinguishing quality of electrocardiosignal | |
| CN102779234A (en) | Electrocardiogram classified processing method and device | |
| CN101449973A (en) | Judgment index generation method and device for cardiac interference signal identification | |
| CN108693448B (en) | Partial discharge mode recognition system applied to power equipment | |
| CN107422266A (en) | A kind of method for diagnosing faults and device of high capacity cell energy-storage system | |
| CN102279358A (en) | MCSKPCA based neural network fault diagnosis method for analog circuits | |
| CN101987017A (en) | Electroencephalo-graph (EEG) signal identification and detection method for measuring alertness of driver | |
| CN105550653A (en) | ECG lead intelligent selection method and system | |
| CN106073712A (en) | Driving based on heart physiological signal warning direction indicators cover device and signal detecting method | |
| CN103439631A (en) | Method and system for detecting corrosion state of grounding grid | |
| CN106913351A (en) | A kind of recognition methods of Mental Workload level | |
| CN103671190A (en) | Intelligent early stage on-line fault diagnosis system of mine fan | |
| CN102068249B (en) | Maternal fetus monitoring device and method | |
| CN102258368A (en) | Time-domain sparsity linear aliasing blind separation model discrimination method in fetal electrocardiogram detection | |
| CN109567832A (en) | A kind of method and system of the angry driving condition of detection based on Intelligent bracelet | |
| CN106889987A (en) | Uterine myoelectricity strength information extracting method based on region filtering treatment | |
| CN113283310A (en) | System and method for detecting health state of power equipment based on voiceprint features |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |