CN107958213A - A kind of cospace pattern based on the medical treatment of brain-computer interface recovering aid and deep learning method - Google Patents
A kind of cospace pattern based on the medical treatment of brain-computer interface recovering aid and deep learning method Download PDFInfo
- Publication number
- CN107958213A CN107958213A CN201711159910.4A CN201711159910A CN107958213A CN 107958213 A CN107958213 A CN 107958213A CN 201711159910 A CN201711159910 A CN 201711159910A CN 107958213 A CN107958213 A CN 107958213A
- Authority
- CN
- China
- Prior art keywords
- msub
- layer
- mrow
- mover
- classification
- 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.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 38
- 238000013135 deep learning Methods 0.000 title claims abstract description 15
- 230000003340 mental effect Effects 0.000 claims abstract description 32
- 238000000605 extraction Methods 0.000 claims abstract description 16
- 230000008569 process Effects 0.000 claims abstract description 16
- 238000001914 filtration Methods 0.000 claims abstract description 13
- 239000011159 matrix material Substances 0.000 claims description 33
- 238000012549 training Methods 0.000 claims description 26
- 230000006870 function Effects 0.000 claims description 20
- 238000000354 decomposition reaction Methods 0.000 claims description 9
- 230000004913 activation Effects 0.000 claims description 5
- 238000006243 chemical reaction Methods 0.000 claims description 4
- 230000009466 transformation Effects 0.000 claims description 4
- 238000000844 transformation Methods 0.000 claims description 4
- 238000013461 design Methods 0.000 claims description 3
- 238000002474 experimental method Methods 0.000 claims description 3
- 238000012880 independent component analysis Methods 0.000 claims description 3
- 210000003141 lower extremity Anatomy 0.000 claims description 3
- 210000001699 lower leg Anatomy 0.000 claims description 3
- 238000002156 mixing Methods 0.000 claims description 3
- 230000000644 propagated effect Effects 0.000 claims description 3
- 238000013139 quantization Methods 0.000 claims description 3
- 238000005070 sampling Methods 0.000 claims description 3
- 210000001364 upper extremity Anatomy 0.000 claims description 3
- 230000002087 whitening effect Effects 0.000 claims description 3
- 230000033001 locomotion Effects 0.000 abstract description 11
- 238000013527 convolutional neural network Methods 0.000 description 19
- 208000006011 Stroke Diseases 0.000 description 13
- 210000004556 brain Anatomy 0.000 description 9
- 230000002490 cerebral effect Effects 0.000 description 7
- 206010008190 Cerebrovascular accident Diseases 0.000 description 6
- 238000011156 evaluation Methods 0.000 description 6
- 230000000694 effects Effects 0.000 description 5
- 238000005516 engineering process Methods 0.000 description 5
- 238000011160 research Methods 0.000 description 5
- 238000010586 diagram Methods 0.000 description 4
- 210000003414 extremity Anatomy 0.000 description 3
- 230000036541 health Effects 0.000 description 3
- 238000013507 mapping Methods 0.000 description 3
- 230000007659 motor function Effects 0.000 description 3
- 208000012661 Dyskinesia Diseases 0.000 description 2
- 206010033799 Paralysis Diseases 0.000 description 2
- 238000012512 characterization method Methods 0.000 description 2
- 230000008034 disappearance Effects 0.000 description 2
- 230000002708 enhancing effect Effects 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- 210000005036 nerve Anatomy 0.000 description 2
- 230000001537 neural effect Effects 0.000 description 2
- 238000005457 optimization Methods 0.000 description 2
- 238000011084 recovery Methods 0.000 description 2
- 208000014644 Brain disease Diseases 0.000 description 1
- 208000034657 Convalescence Diseases 0.000 description 1
- 241001269238 Data Species 0.000 description 1
- 206010020772 Hypertension Diseases 0.000 description 1
- 208000001738 Nervous System Trauma Diseases 0.000 description 1
- 230000009471 action Effects 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000000386 athletic effect Effects 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 239000008280 blood Substances 0.000 description 1
- 210000004369 blood Anatomy 0.000 description 1
- 230000008344 brain blood flow Effects 0.000 description 1
- 210000005013 brain tissue Anatomy 0.000 description 1
- 230000003130 cardiopathic effect Effects 0.000 description 1
- 230000006378 damage Effects 0.000 description 1
- 230000006735 deficit Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 1
- 208000035475 disorder Diseases 0.000 description 1
- 239000003814 drug Substances 0.000 description 1
- 230000009977 dual effect Effects 0.000 description 1
- 230000004064 dysfunction Effects 0.000 description 1
- 235000013399 edible fruits Nutrition 0.000 description 1
- 230000005611 electricity Effects 0.000 description 1
- 208000019622 heart disease Diseases 0.000 description 1
- 230000001771 impaired effect Effects 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 230000008111 motor development Effects 0.000 description 1
- 230000007433 nerve pathway Effects 0.000 description 1
- 208000028412 nervous system injury Diseases 0.000 description 1
- 230000036403 neuro physiology Effects 0.000 description 1
- 230000007935 neutral effect Effects 0.000 description 1
- 230000000474 nursing effect Effects 0.000 description 1
- 208000021090 palsy Diseases 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 230000011514 reflex Effects 0.000 description 1
- 230000008439 repair process Effects 0.000 description 1
- 230000000630 rising effect Effects 0.000 description 1
- 210000004761 scalp Anatomy 0.000 description 1
- 238000001228 spectrum Methods 0.000 description 1
- 230000002195 synergetic effect Effects 0.000 description 1
- 238000010998 test method Methods 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
- 238000013526 transfer learning Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/08—Feature extraction
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/12—Classification; Matching
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Biophysics (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Computational Linguistics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Signal Processing (AREA)
- Image Analysis (AREA)
Abstract
The present invention discloses a kind of cospace pattern based on the medical treatment of brain-computer interface recovering aid and deep learning method, comprises the following steps：Step 1：EEG signals pre-process, and obtain filtering out the EEG signals after noise；Step 2：Improved OVR CSP algorithms can carry out feature extraction to filtering out the multiclass Mental imagery EEG signal after noise, obtain the feature of every a kind of Mental imagery EEG signals, form one-dimensional characteristic data, while the input using variance as grader；Step 3：Further Feature Extraction is carried out using the improved CNN networks for adapting to one-dimensional input sample and is classified.Technical solution using the present invention, more accurately differentiates the movement position and limb motion state of patient, objective data supporting is provided for Measuring scale assessing sufferer rehabilitation degree.
Description
Technical field
The invention belongs to rehabilitation nerve subject technology field, it is related to a kind of based on the medical treatment of brain-computer interface recovering aid
Cospace pattern and deep learning method.
Background technology
Cerebral apoplexy (is commonly called as:Headstroke Stroke), it is due to that cerebral vessels rupture or because angiemphraxis causes blood suddenly
It cannot flow into brain and cause a kind of common brain blood circulation disorder disease of brain tissue impairment.Due to its incidence compared with
Height, has high disability rate, serious threat human health.American Heart Association (AHA) heart disease in 2016 is counted with palsy
Data update shows that cerebral apoplexy is to be only second to cardiopathic global human Health Killer.And China's cerebral apoplexy incidence just
Risen with annual 8.7% speed, illness rate and the death rate are only second to hypertension positioned at second, are brought to society and family heavy
Stress and huge financial burden.
Currently from treatment means analyze, domestic and international rehabilitation of stroke patients system focus primarily on clinical rehabilitation training system with.
Clinical rehabilitation training system uses the rehabilitation training based on neuro-physiology basis more, mainly according to motor development control principle
With brain plasticity principle, using the nervimotion mechanism such as associated movement, synergistic effect and attitudinal reflex, by doctor to patient
Clinical rehabilitation evaluation, judges the functional status and potential ability of patient, then " suits the remedy to the case ", carry out corresponding health
Refreshment is practiced.Method at present clinically on cerebral apoplexy motor functional evaluation is very much, as skeleton symbol Fugl-Meyer motor functions are commented
Point-score, Brunnstrom grade evaluation and test methods etc..The method of these Measuring scale assessings, all relies on the inspection and observation of doctor, belongs to
Manually evaluation, although clinically widely using, evaluation result is easily influenced be subject to physiatrician's subjective factor, and scale score
Level index is more, it is necessary to physiatrician participates in the overall process, and limited doctor's quantity in face of huge sufferer colony often power not from
The heart, or even delay best occasion for the treatment.The present situation in China is clinical rehabilitation resource (such as physiatrician, therapist, nursing at present
Personnel, bed etc.) it is more nervous, and there are serious regional imbalance, and for nervous system injury, medical field does not have also at present
Have the ability to repair completely.The dyskinesia after being ill of stroke patient how is solved the problems, such as by initiative rehabilitation training, is current
The research hotspot and difficult point of medicine.
Thus the help of research high-performance rehabilitation of stroke patients system brain-computer interface can not be complete by the patient of normal output channel
Trained into initiative rehabilitation, so that improving motion function recovers normal activity behavior, for China's Treatment of Cerebral Stroke in neural subject
There is highly important research significance and development prospect with facing Information Science crossing domain.
Existing medical research shows that most cerebral apoplexies, brain and limbs nerve pathway paralysed patient are undamaged, therefore
BCI based on Mental imagery can be used for the stroke area for rebuilding damage, i.e. it is anti-using Mental imagery and nerve that BCI recovers function
The impaired exercise control of feedback, the study that enhancing Motion Control Network is rebuild.
Wherein cospace pattern (Common Spatial Pattern, CSP) is proved to be one of most efficient method, its
Other EEG signals being tested are incorporated into CSP learning processes by the thought of transfer learning, ensure that the brain telecommunications of subject
The estimated bias of number covariance is preferable, is widely used in small training sample data.But with the increase of training sample, it is classified
Accuracy rate increasess slowly；And with the rising of time complexity, the practical application of limit algorithm.
In recent years, go deep into the research of deep learning, many brain electricity sample datas can be incorporated into such as convolutional Neural
Processing is trained in network frame, it is necessary to relatively great amount of training sample.The convolution number of plies can be chosen according to sample size, and one
Denier sample size data are too small, cause identification error rate to greatly promote.For multiclass Mental imagery EEG signals, sample size is generally
Small Amount, and quoting deep learning frame merely cannot train up, it is difficult to play the advantage of CNN algorithms.Therefore, it is of the invention
It is proposed to classify the mode that improved CSP algorithms and CNN are combined to multiclass Mental imagery EEG signals.
In conclusion using the Mental imagery EEG signals of brain-computer interface technology identification patient, can be by patient motion wish
Translate into control command and drive convalescence device action, help patient to complete initiative rehabilitation training, it is extensive to be conducive to improvement motor function
Multiple effect, and do not tending to be ripe also using upper for the algorithm of this technology.
The content of the invention
The present invention provides a kind of cospace pattern based on the medical treatment of brain-computer interface recovering aid and deep learning method, for
How the rehabilitation training field of study movement dysfunction, realize high accuracy, the brain based on Mental imagery of high robust
Machine interfacing, so as to accurately differentiate athletic posture and the limb motion position of patient, carries for Measuring scale assessing sufferer rehabilitation degree
For objective data supporting, effectively alleviate current Treatment of Cerebral Stroke equipment and doctor's resource wretched insufficiency, what subjective treatment zone was come
The problems such as drawback.The pretreatment that EEG signals are carried out using wavelet package transforms and quick independent analytical methods is proposed, mainly
Original Mental imagery EEG signals are filtered, the various noises being reduced as far as in EEG signals, improve signal-to-noise ratio,
Based on improved cospace pattern combination convolutional neural networks algorithm identification patient motion imagination EEG signals, so that more accurately
Differentiate the movement position and limb motion state of patient, objective data supporting is provided for Measuring scale assessing sufferer rehabilitation degree.
To achieve the above object, the present invention adopts the following technical scheme that：
A kind of cospace pattern based on the medical treatment of brain-computer interface recovering aid and deep learning method, comprise the following steps：
Step 1：EEG signals pre-process, and obtain filtering out the EEG signals after noise；
Step 2：Improved OVR-CSP algorithms can carry out feature to filtering out the multiclass Mental imagery EEG signal after noise
Extraction, obtains the feature of every a kind of Mental imagery EEG signals, forms one-dimensional characteristic data, while using variance as grader
Input；
Step 3：Further Feature Extraction is carried out using the improved CNN networks for adapting to one-dimensional input sample and is classified.
Preferably, step 1 specifically includes：Two processes of wavelet package transforms and Fast Independent Component Analysis：
WPT is carried out to original EEG first, determines the Decomposition order of WPT, it is suitable to be selected according to the characteristics of EEG and noise
Wavelet basis function, finally determine to filter out frequency band where high-frequency noise, and to corresponding frequency band zero setting；
FastICA conversion is carried out to oneself signal after filtering out high-frequency noise again, FastICA inverse transformations and FastICA are inverse
Conversion, obtains filtering out the EEG signals after noise.
Preferably, step 2 is specially：OVR-CSP is that five type games imagination task is divided into five two new classes to classify
Problem, obtains five projection matrixes, and five groups of corresponding spatial features can be obtained after projection；
Specific calculating process is as follows:If Xi(i=1,2,3,4,5) is the N*T EEG signals of five generic tasks, and N believes for collection
Number channel number, T be each passage sampling number, T >=N, the basic principle of OVR-CSP algorithms is to calculate five classes respectively
The normalized covariance matrix R of datai, i.e.,：
Blending space covariance matrix R, which can be obtained, is：
Wherein,For the average covariance matrices of five generic task many experiments, characteristic value point is carried out to R
Solution, obtains：
R=UVUT
Wherein, U is the feature vector of R, and V is the eigenvalue matrix of R；Descending arrangement is carried out to eigenvalue matrix, according to row
Feature vector is made identical adjustment in position after sequence, then whitening matrix P is：
P=V-1/2UT
OVR-CSP is classified as one kind when calculating projection matrix, by one type, and remaining four class is then to be another kind of, i.e.,：
It is X respectively by two new classes are divided into by the EEG signals X of pretreatment1,X1', the projection under the i-th quasi-mode
Direction is projected, and is obtained：
Z1=(U1′)TP1X1,Z1'=(U1′)TP1X1' (i=1,2,3,4,5)
The covariance matrix value of matrix of the five class data after projection is
The feature vector of covariance matrix is normalized again, is obtained：
Wherein, n be feature vector columns, in this, as feature vector into
Row classification learning.
Preferably, step 3 is specially：For the CNN structures of Mental imagery eeg signal classification, MIEEGNet is named as
Network structure, the network are formed by 5 layers, are input layer including first layer, two layers of convolutional layer, 1 layer of full articulamentum and 1 layer
Softmax classification layers；Sample data size is 1*fi,j；Wherein, N is Mental imagery EEG through being obtained after OVR-CSP algorithm process
The Characteristic Number arrived, fi,j=5*m；The second layer (C2) and third layer (C3) are all convolutional layer, to be carried out to input sample data
Further Feature Extraction, the wherein second layer (C2 } there is i2A convolution kernel, the size of convolution kernel is 1*n2, third layer (C3) has i3A volume
Product core, the size of convolution kernel is 1*n3；4th layer (F4) is full articulamentum, its by way of connecting entirely with layer 5 (O5) one
Rise and form individual layer perceptron, by output category result after third layer (C3) output result treatment.
MIEEGNet networks are classified in last layer using Softmax functions, give input fi,jWhen, the input number
According to be c classification probability be denoted as p (y=c | fi,j), then network losses function Loss is defined as follows,
Wherein, fi,jFor the Mental imagery feature training sample of input, i.e., the covariance matrix after OVR-CSP is normalized
Feature vector；Y is the corresponding class label of sample (having determined that classification)；f(fi,j) it is to be exported by the network of MIEEGNet；
Then output layer starts the weights of convolution kernel in backpropagation adjustment network；Finally pass through the training of MIEEGNet, export Loss
Value；f(fi,j) output be a left side, the right hand, tongue is left, the classification of right crus of diaphragm；And controlled the classification results of Mental imagery as BCI
Pass down the line processed gives mechanical upper and lower extremities, as quantization assessment index.
Preferably, the network training includes two stages：
The propagated forward stage：From training sample set sample drawn fi,j, its corresponding class label is y, and x is input to this
The MIEEGNet networks of Subject Design, the output of last layer are exactly the input of current layer, are calculated and worked as by activation primitive ReLU
The output of front layer, then successively hands on.Finally by classification layer outputFor three-dimensional feature vector, its element represents
Sample data fi,jBelong to the probability of classification c；
Back-propagation phase：Calculate the output of classification layerThe error of class label y is given with sample, and uses minimum
The method adjustment weighting parameter of mean square error cost function.
Brief description of the drawings
Fig. 1 Technology Roadmaps proposed by the present invention
Fig. 2 brain computator method schematic diagrames proposed by the present invention based on Mental imagery.
The EEG signals WAVELET PACKET DECOMPOSITION schematic diagram of Fig. 3 present invention.
The pretreatment schematic diagram of the EEG signals of Fig. 4 present invention.
Fig. 5 FB(flow block)s of the present invention.
Fig. 6 MIEEGNet network structures proposed by the present invention.
Embodiment
As shown in Fig. 2, a kind of cospace pattern based on the medical treatment of brain-computer interface recovering aid of the invention and deep learning side
Method, CSP is combined and is improved with two kinds of algorithms of CNN, and Mental imagery EEG signals are carried out with Further Feature Extraction and is divided
Class.Compared with directly inputting original EEG signals, the discrimination between signal is not only increased, and makes CNN input sample numbers
It is one-dimensional according to being fallen below by two dimension, input sample size of data is greatly reduced, reduces the number of convolution kernel in network, makes needs
Trained network weight number substantially reduces.
As shown in Figure 1, a kind of cospace pattern based on the medical treatment of brain-computer interface recovering aid and deep learning method, including
Following steps：
Step 1：EEG signals pre-process, including two processes of wavelet package transforms and Fast Independent Component Analysis：
First, WPT is carried out to original Mental imagery brain electric information, determines the Decomposition order of WPT, utilize wavelet package transforms
When filtering out the high-frequency noise in EEG signals, it is contemplated that the enhancing of the power spectrum of characterization EEG signals ERD/ERS phenomenons or decrease
Frequency range is mainly reflected in 8~30Hz, is that 128Hz can determine that the Decomposition order of WPT is five layers for sample frequency therefore,
Determine frequency range in 8~30Hz by WAVELET PACKET DECOMPOSITION.Finally determine to filter out the frequency band where high-frequency noise, and to corresponding
Frequency band zero setting, if Fig. 3 is WAVELET PACKET DECOMPOSITION schematic diagram.
Then, FastICA conversion then to oneself signal after filtering out high-frequency noise is carried out, obtains the EEG signals of each passage
Independent element, then calculate the C3 of each independent element and original EEG signal respectively, the related coefficient of C4 passages, retains phase
Relation number is higher than the independent element of the threshold value of setting, then FastICA inverse transformations, obtains filtering out the EEG signals after noise, such as Fig. 4
It is shown.Two kinds of algorithms of WPT and FastICA pre-process EEG signals, and the noise and part for having filtered out EEG signals disturb
Signal, easy to follow-up feature extraction and classification.
Step 2:Feature extraction is carried out to multiclass Mental imagery EEG signal using OVR-CSP algorithms.
Assuming that the number for the projection matrix that W classes signal needs to try to achieve is W, W groups spatial domain knot is can obtain after each sample projection
Fruit, which is the feature per a kind of Mental imagery EEG signals extracted, finally using variance as the defeated of grader
Enter.Feature after extraction can form one-dimensional characteristic data.It enormously simplify convolutional neural networks structure：Both can increase each
Discrimination between EEG signal, can also reduce the size of data of convolutional neural networks input sample.It is illustrated in figure 5 and is based on
Five type games imagination task (being left hand, the right hand, foot and tongue respectively) feature extraction of OVR-CSP.
OVR-CSP is that five type games imagination task is divided into five two new class classification problems, obtains five projection matrixes,
Five groups of corresponding spatial features can be obtained after projection.Specific calculating process is as follows:If Xi(i=1,2,3,4,5) appoint for five classes
The N*T EEG signals of business, N be collection signal channel number, T be each passage sampling number, T >=N.OVR-CSP algorithms
Basic principle be to calculate the normalized covariance matrix R of five class data respectivelyi, i.e.,：
Blending space covariance matrix R, which can be obtained, is：
WhereinFor the average covariance matrices of five generic task many experiments.Characteristic value point is carried out to R
Solution, obtains：
R=UVUT
Wherein, U is the feature vector of R；V is the eigenvalue matrix of R.Descending arrangement is carried out to eigenvalue matrix, according to row
Feature vector is made identical adjustment in position after sequence, then whitening matrix P is：
P=V-1/2UT
Unlike two classical class CSP algorithms, one type is classified as one by OVR-CSP when calculating projection matrix
Class, and remaining four class is then to be another kind of, i.e.,：
It is X respectively by two new classes are divided into by the EEG signals X of pretreatment1,X1', the projection under the i-th quasi-mode
Direction is projected, and is obtained：
Z1=(U1′)TP1X1,Z1'=(U1′)TP1X1' (i=1,2,3,4,5)
The covariance matrix value of matrix of the five class data after projection is RZi=ZiZi T(i=1,2,3,4,5), then to association
The feature vector of variance matrix is normalized, and obtains：
(n is the columns of feature vector)
Classification learning is carried out in this, as feature vector.
Step 3：Further Feature Extraction is carried out using the improved CNN networks for adapting to one-dimensional input sample and is classified.
The present invention devises a kind of CNN structures for Mental imagery eeg signal classification, is named as MIEEGNet networks
Structure, CNN network structures such as Fig. 6.The network is formed by 5 layers, is input layer including first layer, two layers of convolutional layer, 1 layer it is complete
Articulamentum and 1 layer of Softmax classification layer.Sample data size is 1*fi,j.Wherein, N is Mental imagery EEG through OVR-CSP algorithms
Handle the Characteristic Number obtained afterwards, fi,j=5*m；The second layer (C2) and third layer (C3) are all convolutional layer, to inputting sample
Notebook data progress Further Feature Extraction, the wherein second layer (C2 } there is i2A convolution kernel, the size of convolution kernel is 1*n2, third layer
(C3) there is i3A convolution kernel, the size of convolution kernel is 1*n3.Since input sample data length is less than normal, so CNN networks of the present invention
Down-sampled layer is saved, ReLU function optimization parameters is used in convolutional layer, avoids " gradient disappearance "；4th layer (F4) is full connection
Layer, it forms individual layer perceptron by way of connecting entirely together with layer 5 (O5), and third layer (C3) is exported result treatment
Output category result afterwards.
MIEEGNet networks are classified in last layer using Softmax functions.The given input f of the present inventioni,jWhen, should
Input data be c classification probability be denoted as p (y=c | fi,j), then network losses function Loss is defined as follows,
Wherein fi,jFor the Mental imagery feature training sample of input, i.e., the covariance matrix after OVR-CSP is normalized
Feature vector；Y is the corresponding class label of sample (having determined that classification)；f(fi,j) it is to be exported by the network of MIEEGNet.So
Output layer starts the weights of convolution kernel in backpropagation adjustment network afterwards.Finally pass through the training of MIEEGNet, export Loss's
Value, the value of Loss is smaller, illustrates that training effect is better, classification results are more accurate.f(fi,j) output for a left side, the right hand, tongue,
A left side, the classification of right crus of diaphragm.And mechanical upper and lower extremities are communicated to using the classification results of Mental imagery as BCI control commands, as quantization
Deliberated index.Compared to only with the experience deliberated index of doctor, there is provided objective data supporting, feeds back to physiatrician.Rehabilitation
Doctor carries out constantly renewal evaluation and test model again, and then as the new method of guiding treatment, is carried out for patients with cerebral apoplexy effective
Rehabilitation training, until helping patient's early recovery.
The number of filter of each layer traditional of CNN and its modified network is all fixed constant, but and not all filter
Ripple device, which can select the other implicit features of target class, to be come.Since the number and number of filter of CNN network parameters are close
Correlation, therefore in back-propagation phase, can be according to that minimizes mean square error cost function and adaptively adjust each layer of wave filter
Number (convolution kernel number), so as to reduce algorithm complex while accuracy of identification is ensured.This problem introduces convolution kernel number and becomes
K is measured, convolutional layer C2, C3 are respectively adopted 2kWith 2k+1A convolution kernel carries out convolution operation to every layer of input.
Activation primitive of the sigmoid functions as convolutional network, this hair are used in traditional CNN network characterization mapping structures
Bright use instead corrects linear unit (Rectified Linear Units, ReLU) function to possess the network after training appropriate dilute
Property is dredged, while the issuable gradient of conventional activation function during backpropagation arameter optimization can be solved well disappears to ask
Topic, accelerates the convergence of network.
Network structure proposed by the present invention realizes the mapping relations for being input to output, efficiently solves due to traditional convolution
The gradient disappearance problem of neutral net.
Wherein network training includes two stages：The propagated forward stage.From training sample set sample drawn fi,j, it is corresponded to
Class label be y.X is input to the MIEEGNet networks of this Subject Design, the output of last layer is exactly the input of current layer.
The output of current layer is calculated by activation primitive ReLU, is then successively handed on.Finally by classification layer output
For three-dimensional feature vector, its element representative sample data fi,jBelong to the probability of classification c.Back-propagation phase.Calculate classification layer
OutputThe error of class label y is given with sample, and is joined using the method adjustment weights for minimizing mean square error cost function
Number.
During scalp EEG signals EEG feature extractions, signal decomposition is by the thought present invention introduces patches
Small data slot, forms covariance matrix, then by Matrix Estimation extraction Mental imagery EEG signals based on OVR-CSP side
The characteristic of division vector of method；After high dimensional feature DUAL PROBLEMS OF VECTOR MAPPING to lower dimensional space, according to minimum mean square error cost function certainly
Adapt to adjust traditional each layer of convolution kernel number of CNN, and the convergence corrected linear unit function using using and accelerate network,
Improve classification effectiveness.
In conclusion the present invention has made the improvement of cospace pattern and convolutional neural networks based on Mental imagery, i.e., gram
It is bad in more classification task expression effects traditional C/S P algorithms are taken, and overcome single use convolutional neural networks to need largely
The two, is cleverly combined by the shortcomings that data sample, and robustness is preferable on small sample in processes.In view of both at home and abroad to brain
The mode of Stroke Rehabilitation treatment is more single.Therefore the rehabilitation training field applied to dyskinesia patient, helps patient
Initiative rehabilitation training is participated in, is conducive to improve motor function recovery effect, it is objective to provide for Measuring scale assessing sufferer rehabilitation degree
Data supporting.
Claims (5)
1. a kind of cospace pattern based on the medical treatment of brain-computer interface recovering aid and deep learning method, it is characterised in that including
Following steps：
Step 1：EEG signals pre-process, and obtain filtering out the EEG signals after noise；
Step 2：Improved OVR-CSP algorithms can carry out feature extraction to filtering out the multiclass Mental imagery EEG signal after noise,
The feature of every a kind of Mental imagery EEG signals is obtained, forms one-dimensional characteristic data, while using variance as the defeated of grader
Enter；
Step 3：Further Feature Extraction is carried out using the improved CNN networks for adapting to one-dimensional input sample and is classified.
2. cospace pattern and deep learning method as claimed in claim 1 based on the medical treatment of brain-computer interface recovering aid, its
It is characterized in that, step 1 specifically includes：Two processes of wavelet package transforms and Fast Independent Component Analysis：
WPT is carried out to original EEG first, determines the Decomposition order of WPT, is selected according to the characteristics of EEG and noise suitable small
Ripple basic function, finally determines to filter out the frequency band where high-frequency noise, and to corresponding frequency band zero setting；
FastICA conversion, FastICA inverse transformations and FastICA inverse transformations are carried out to oneself signal after filtering out high-frequency noise again,
Obtain filtering out the EEG signals after noise.
3. cospace pattern and deep learning method as claimed in claim 2 based on the medical treatment of brain-computer interface recovering aid, its
It is characterized in that step 2 is specially：OVR-CSP is that five type games imagination task is divided into five two new class classification problems, is obtained
Five projection matrixes, can obtain five groups of corresponding spatial features after projection；
Specific calculating process is as follows:If Xi(i=1,2,3,4,5) is the N*T EEG signals of five generic tasks, and N is the logical of collection signal
Road number, T are the sampling number of each passage, and T >=N, the basic principle of OVR-CSP algorithms is to calculate five class data respectively
Normalized covariance matrix Ri, i.e.,：
<mrow>
<msub>
<mi>R</mi>
<mi>i</mi>
</msub>
<mo>=</mo>
<mfrac>
<mrow>
<msub>
<mi>X</mi>
<mi>i</mi>
</msub>
<msup>
<msub>
<mi>X</mi>
<mi>i</mi>
</msub>
<mi>T</mi>
</msup>
</mrow>
<mrow>
<mi>t</mi>
<mi>r</mi>
<mi>a</mi>
<mi>c</mi>
<mi>e</mi>
<mrow>
<mo>(</mo>
<msub>
<mi>X</mi>
<mi>i</mi>
</msub>
<msup>
<msub>
<mi>X</mi>
<mi>i</mi>
</msub>
<mi>T</mi>
</msup>
<mo>)</mo>
</mrow>
</mrow>
</mfrac>
<mo>,</mo>
<mi>i</mi>
<mo>=</mo>
<mn>1</mn>
<mo>,</mo>
<mn>2</mn>
<mo>,</mo>
<mn>3</mn>
<mo>,</mo>
<mn>4</mn>
<mo>,</mo>
<mn>5</mn>
</mrow>
Blending space covariance matrix R, which can be obtained, is：
<mrow>
<mi>R</mi>
<mo>=</mo>
<mover>
<msub>
<mi>R</mi>
<mn>1</mn>
</msub>
<mo>&OverBar;</mo>
</mover>
<mo>+</mo>
<mover>
<msub>
<mi>R</mi>
<mn>2</mn>
</msub>
<mo>&OverBar;</mo>
</mover>
<mo>+</mo>
<mover>
<msub>
<mi>R</mi>
<mn>3</mn>
</msub>
<mo>&OverBar;</mo>
</mover>
<mo>+</mo>
<mover>
<msub>
<mi>R</mi>
<mn>4</mn>
</msub>
<mo>&OverBar;</mo>
</mover>
<mo>+</mo>
<mover>
<msub>
<mi>R</mi>
<mn>5</mn>
</msub>
<mo>&OverBar;</mo>
</mover>
</mrow>
Wherein,For the average covariance matrices of five generic task many experiments, Eigenvalues Decomposition is carried out to R,
：
R=UVUT
Wherein, U is the feature vector of R, and V is the eigenvalue matrix of R；Descending arrangement is carried out to eigenvalue matrix, after sequence
Position make identical adjustment to feature vector, then whitening matrix P is：
P=V-1/2UT
OVR-CSP is classified as one kind when calculating projection matrix, by one type, and remaining four class is then to be another kind of, i.e.,：
<mrow>
<msup>
<msub>
<mi>R</mi>
<mn>1</mn>
</msub>
<mo>&prime;</mo>
</msup>
<mo>=</mo>
<mover>
<msub>
<mi>R</mi>
<mn>2</mn>
</msub>
<mo>&OverBar;</mo>
</mover>
<mo>+</mo>
<mover>
<msub>
<mi>R</mi>
<mn>3</mn>
</msub>
<mo>&OverBar;</mo>
</mover>
<mo>+</mo>
<mover>
<msub>
<mi>R</mi>
<mn>4</mn>
</msub>
<mo>&OverBar;</mo>
</mover>
<mo>+</mo>
<mover>
<msub>
<mi>R</mi>
<mn>5</mn>
</msub>
<mo>&OverBar;</mo>
</mover>
</mrow>
It is X respectively by two new classes are divided into by the EEG signals X of pretreatment1,X1', the projecting direction under the i-th quasi-mode
Projected, obtained：
Z1=(U1′)TP1X1,Z1'=(U1′)TP1X1' (i=1,2,3,4,5)
The covariance matrix value of matrix of the five class data after projection is
The feature vector of covariance matrix is normalized again, is obtained：
Wherein, n is the columns of feature vector, is divided in this, as feature vector
Class learns.
4. cospace pattern and deep learning method as claimed in claim 3 based on the medical treatment of brain-computer interface recovering aid, its
It is characterized in that step 3 is specially：For the CNN structures of Mental imagery eeg signal classification, MIEEGNet network structures are named as,
The network is formed by 5 layers, is input layer including first layer, and two layers of convolutional layer, 1 layer of full articulamentum and 1 layer of Softmax classify
Layer；Sample data size is 1*fi,j；Wherein, N is the Mental imagery EEG features through being obtained after OVR-CSP algorithm process
Number, fi,j=5*m；The second layer (C2) and third layer (C3) are all convolutional layer, are carried to carry out quadratic character to input sample data
Take, wherein the second layer (C2 } have i2A convolution kernel, the size of convolution kernel is 1*n2, third layer (C3) has i3A convolution kernel, convolution kernel
Size be 1*n3；4th layer (F4) is full articulamentum, it forms individual layer by way of connecting entirely with together with layer 5 (O5)
Perceptron, by output category result after third layer (C3) output result treatment.
MIEEGNet networks are classified in last layer using Softmax functions, give input fi,jWhen, which is c
The probability of classification be denoted as p (y=c | fi,j), then network losses function Loss is defined as follows,
<mrow>
<mi>L</mi>
<mi>o</mi>
<mi>s</mi>
<mi>s</mi>
<mrow>
<mo>(</mo>
<mi>f</mi>
<mo>(</mo>
<msub>
<mi>f</mi>
<mrow>
<mi>i</mi>
<mo>,</mo>
<mi>j</mi>
</mrow>
</msub>
<mo>)</mo>
<mo>,</mo>
<mi>y</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mo>-</mo>
<msub>
<mi>&Sigma;</mi>
<mi>c</mi>
</msub>
<mi>log</mi>
<mi>p</mi>
<mrow>
<mo>(</mo>
<mi>y</mi>
<mo>=</mo>
<mi>c</mi>
<mo>|</mo>
<msub>
<mi>f</mi>
<mrow>
<mi>i</mi>
<mo>,</mo>
<mi>j</mi>
</mrow>
</msub>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mo>-</mo>
<mi>log</mi>
<mi>p</mi>
<mrow>
<mo>(</mo>
<mi>y</mi>
<mo>=</mo>
<mover>
<mi>y</mi>
<mo>~</mo>
</mover>
<mo>|</mo>
<mi>x</mi>
<mo>)</mo>
</mrow>
</mrow>
Wherein, fi,jFor the Mental imagery feature training sample of input, i.e., the spy of covariance matrix after OVR-CSP is normalized
Sign vector；Y is the corresponding class label of sample (having determined that classification)；f(fi,j) it is to be exported by the network of MIEEGNet；Then
Output layer starts the weights of convolution kernel in backpropagation adjustment network；Finally pass through the training of MIEEGNet, export the value of Loss；
f(fi,j) output be a left side, the right hand, tongue is left, the classification of right crus of diaphragm；And the classification results of Mental imagery are controlled as BCI and are ordered
Order is communicated to mechanical upper and lower extremities, as quantization assessment index.
5. cospace pattern and deep learning method as claimed in claim 4 based on the medical treatment of brain-computer interface recovering aid, its
It is characterized in that, the network training includes two stages：
The propagated forward stage：From training sample set sample drawn fi,j, its corresponding class label is y, and x is input to this problem
The MIEEGNet networks of design, the output of last layer are exactly the input of current layer, and current layer is calculated by activation primitive ReLU
Output, then successively hand on.Finally by classification layer output For three-dimensional feature vector, its element representative sample number
According to fi,jBelong to the probability of classification c；
Back-propagation phase：Calculate the output of classification layerThe error of class label y is given with sample, and it is square using minimizing
The method adjustment weighting parameter of error cost function.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711159910.4A CN107958213A (en) | 2017-11-20 | 2017-11-20 | A kind of cospace pattern based on the medical treatment of brain-computer interface recovering aid and deep learning method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711159910.4A CN107958213A (en) | 2017-11-20 | 2017-11-20 | A kind of cospace pattern based on the medical treatment of brain-computer interface recovering aid and deep learning method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN107958213A true CN107958213A (en) | 2018-04-24 |
Family
ID=61963963
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201711159910.4A Pending CN107958213A (en) | 2017-11-20 | 2017-11-20 | A kind of cospace pattern based on the medical treatment of brain-computer interface recovering aid and deep learning method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107958213A (en) |
Cited By (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109003680A (en) * | 2018-09-28 | 2018-12-14 | 四川大学 | Epilepsy data statistical approach and device |
| CN109009092A (en) * | 2018-06-15 | 2018-12-18 | 东华大学 | A method of removal EEG signals noise artefact |
| CN109009097A (en) * | 2018-07-18 | 2018-12-18 | 厦门大学 | A kind of brain electricity classification method of adaptive different sample frequencys |
| CN109144266A (en) * | 2018-08-29 | 2019-01-04 | 安徽大学 | A kind of brain-computer interface lead optimization method based on independent component analysis |
| CN109164910A (en) * | 2018-07-05 | 2019-01-08 | 北京航空航天大学合肥创新研究院 | For the multiple signals neural network architecture design method of electroencephalogram |
| CN109325586A (en) * | 2018-12-05 | 2019-02-12 | 北京航空航天大学合肥创新研究院 | Deep neural network system based on composite object function |
| CN109620223A (en) * | 2018-12-07 | 2019-04-16 | 北京工业大学 | A kind of rehabilitation of stroke patients system brain-computer interface key technology method |
| CN109685031A (en) * | 2018-12-29 | 2019-04-26 | 河海大学常州校区 | A kind of brain-computer interface midbrain signal characteristics classification method and system |
| CN109726751A (en) * | 2018-12-21 | 2019-05-07 | 北京工业大学 | Method based on depth convolutional neural networks identification brain Electrical imaging figure |
| CN110188836A (en) * | 2019-06-21 | 2019-08-30 | 西安交通大学 | A kind of brain function network class method based on variation self-encoding encoder |
| CN110569727A (en) * | 2019-08-06 | 2019-12-13 | 华南理工大学 | Transfer learning method combining intra-class distance and inter-class distance based on motor imagery classification |
| CN110851783A (en) * | 2019-11-12 | 2020-02-28 | 华中科技大学 | Heterogeneous label space migration learning method for brain-computer interface calibration |
| CN111544856A (en) * | 2020-04-30 | 2020-08-18 | 天津大学 | Brain-myoelectricity intelligent full limb rehabilitation method based on novel transfer learning model |
| CN111990992A (en) * | 2020-09-03 | 2020-11-27 | 山东中科先进技术研究院有限公司 | Electroencephalogram-based autonomous movement intention identification method and system |
| CN112185558A (en) * | 2020-09-22 | 2021-01-05 | 珠海中科先进技术研究院有限公司 | Mental health and rehabilitation evaluation method, device and medium based on deep learning |
| CN113128384A (en) * | 2021-04-01 | 2021-07-16 | 北京工业大学 | Brain-computer interface software key technical method of stroke rehabilitation system based on deep learning |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104771163A (en) * | 2015-01-30 | 2015-07-15 | 杭州电子科技大学 | Electroencephalogram feature extraction method based on CSP and R-CSP algorithms |
-
2017
- 2017-11-20 CN CN201711159910.4A patent/CN107958213A/en active Pending
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104771163A (en) * | 2015-01-30 | 2015-07-15 | 杭州电子科技大学 | Electroencephalogram feature extraction method based on CSP and R-CSP algorithms |
Non-Patent Citations (3)
| Title |
|---|
| 曾庆山: ""基于CSP与卷积神经网络算法的多类运动"", 《科学技术与工程》 * |
| 王洪涛: ""混合脑机接口实现及其应用研究"", 《中国博士学位论文全文数据库 医药卫生科技辑》 * |
| 范明莉: ""基于卷积神经网络的运动想象脑电信号特征提取与分类"", 《中国优秀硕士学位论文全文数据库 医药卫生科技辑》 * |
Cited By (24)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109009092B (en) * | 2018-06-15 | 2020-06-02 | 东华大学 | Method for removing noise artifact of electroencephalogram signal |
| CN109009092A (en) * | 2018-06-15 | 2018-12-18 | 东华大学 | A method of removal EEG signals noise artefact |
| CN109164910A (en) * | 2018-07-05 | 2019-01-08 | 北京航空航天大学合肥创新研究院 | For the multiple signals neural network architecture design method of electroencephalogram |
| CN109164910B (en) * | 2018-07-05 | 2021-09-21 | 北京航空航天大学合肥创新研究院 | Multi-signal neural network architecture design method for electroencephalogram |
| CN109009097A (en) * | 2018-07-18 | 2018-12-18 | 厦门大学 | A kind of brain electricity classification method of adaptive different sample frequencys |
| CN109144266A (en) * | 2018-08-29 | 2019-01-04 | 安徽大学 | A kind of brain-computer interface lead optimization method based on independent component analysis |
| CN109003680A (en) * | 2018-09-28 | 2018-12-14 | 四川大学 | Epilepsy data statistical approach and device |
| CN109325586A (en) * | 2018-12-05 | 2019-02-12 | 北京航空航天大学合肥创新研究院 | Deep neural network system based on composite object function |
| CN109325586B (en) * | 2018-12-05 | 2022-03-04 | 北京航空航天大学合肥创新研究院 | System for denoising electroencephalogram signal |
| CN109620223A (en) * | 2018-12-07 | 2019-04-16 | 北京工业大学 | A kind of rehabilitation of stroke patients system brain-computer interface key technology method |
| CN109726751A (en) * | 2018-12-21 | 2019-05-07 | 北京工业大学 | Method based on depth convolutional neural networks identification brain Electrical imaging figure |
| CN109726751B (en) * | 2018-12-21 | 2020-11-27 | 北京工业大学 | Method for recognizing electroencephalogram based on deep convolutional neural network |
| CN109685031B (en) * | 2018-12-29 | 2022-09-13 | 河海大学常州校区 | Electroencephalogram signal feature classification method and system in brain-computer interface |
| CN109685031A (en) * | 2018-12-29 | 2019-04-26 | 河海大学常州校区 | A kind of brain-computer interface midbrain signal characteristics classification method and system |
| CN110188836A (en) * | 2019-06-21 | 2019-08-30 | 西安交通大学 | A kind of brain function network class method based on variation self-encoding encoder |
| CN110188836B (en) * | 2019-06-21 | 2021-06-11 | 西安交通大学 | Brain function network classification method based on variational self-encoder |
| CN110569727A (en) * | 2019-08-06 | 2019-12-13 | 华南理工大学 | Transfer learning method combining intra-class distance and inter-class distance based on motor imagery classification |
| CN110569727B (en) * | 2019-08-06 | 2022-03-29 | 华南理工大学 | Transfer learning method combining intra-class distance and inter-class distance for motor imagery classification |
| CN110851783A (en) * | 2019-11-12 | 2020-02-28 | 华中科技大学 | Heterogeneous label space migration learning method for brain-computer interface calibration |
| CN111544856A (en) * | 2020-04-30 | 2020-08-18 | 天津大学 | Brain-myoelectricity intelligent full limb rehabilitation method based on novel transfer learning model |
| CN111990992A (en) * | 2020-09-03 | 2020-11-27 | 山东中科先进技术研究院有限公司 | Electroencephalogram-based autonomous movement intention identification method and system |
| CN112185558A (en) * | 2020-09-22 | 2021-01-05 | 珠海中科先进技术研究院有限公司 | Mental health and rehabilitation evaluation method, device and medium based on deep learning |
| CN113128384A (en) * | 2021-04-01 | 2021-07-16 | 北京工业大学 | Brain-computer interface software key technical method of stroke rehabilitation system based on deep learning |
| CN113128384B (en) * | 2021-04-01 | 2024-04-05 | 北京工业大学 | Brain-computer interface software key technical method of cerebral apoplexy rehabilitation system based on deep learning |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107958213A (en) | A kind of cospace pattern based on the medical treatment of brain-computer interface recovering aid and deep learning method | |
| CN109620223A (en) | A kind of rehabilitation of stroke patients system brain-computer interface key technology method | |
| CN104091172B (en) | A kind of feature extracting method of Mental imagery EEG signals | |
| CN102722728B (en) | Motion image electroencephalogram classification method based on channel weighting supporting vector | |
| CN109784211A (en) | A kind of Mental imagery Method of EEG signals classification based on deep learning | |
| US20040220782A1 (en) | Signal interpretation engine | |
| CN110619322A (en) | Multi-lead electrocardio abnormal signal identification method and system based on multi-flow convolution cyclic neural network | |
| CN106682616A (en) | Newborn-painful-expression recognition method based on dual-channel-characteristic deep learning | |
| CN105249964B (en) | Appraisal procedure is rebuild based on magneticencephalogram and the multi-modal brain function of diffusion tensor | |
| CN107944493A (en) | A kind of the first lead electrocardiogram beat classification method based on convolutional neural networks | |
| CN113180692B (en) | Electroencephalogram signal classification and identification method based on feature fusion and attention mechanism | |
| CN107811649A (en) | A kind of more sorting techniques of heart sound based on depth convolutional neural networks | |
| CN111544855A (en) | Pure idea control intelligent rehabilitation method based on distillation learning and deep learning and application | |
| CN109498370A (en) | Joint of lower extremity angle prediction technique based on myoelectricity small echo correlation dimension | |
| Rahimian et al. | Hand gesture recognition using temporal convolutions and attention mechanism | |
| CN107480637A (en) | Heart failure based on heart sound feature method by stages | |
| CN115206484B (en) | Cerebral apoplexy rehabilitation training system | |
| Ahsan et al. | EMG motion pattern classification through design and optimization of neural network | |
| Das et al. | CA-Net: A novel cascaded attention-based network for multi-stage glaucoma classification using fundus images | |
| CN103908361A (en) | Method for acquiring and operating artificial limb joint movement coupling drive signals | |
| CN109994202A (en) | A method of the face based on deep learning generates prescriptions of traditional Chinese medicine | |
| CN117153379B (en) | Prediction device for thoracic outlet syndrome | |
| Li et al. | Multi-view fusion network-based gesture recognition using sEMG data | |
| CN116831874A (en) | Lower limb rehabilitation device control method based on electromyographic signals | |
| CN116522106A (en) | Motor imagery electroencephalogram signal classification method based on transfer learning parallel multi-scale filter bank time domain convolution |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| TA01 | Transfer of patent application right | ||
| TA01 | Transfer of patent application right |
Effective date of registration: 20240109 Address after: 100124 No. 100 Chaoyang District Ping Tian Park, Beijing Applicant after: Beijing University of Technology Applicant after: Wang Zhuozheng Address before: 100124 No. 100 Chaoyang District Ping Tian Park, Beijing Applicant before: Beijing University of Technology Effective date of registration: 20240109 Address after: 310007 station 30, 2nd floor, east area, building a, Zhejiang University Science Park, No. 525 Xixi Road, Xihu District, Hangzhou, Zhejiang Province Applicant after: Hangzhou Xingyuan intelligent Biotechnology Co.,Ltd. Address before: 100124 No. 100 Chaoyang District Ping Tian Park, Beijing Applicant before: Beijing University of Technology Applicant before: Wang Zhuozheng |