CN109276244A - The recognition methods that age-care based on brain wave information is intended to - Google Patents
The recognition methods that age-care based on brain wave information is intended to Download PDFInfo
- Publication number
- CN109276244A CN109276244A CN201811022344.7A CN201811022344A CN109276244A CN 109276244 A CN109276244 A CN 109276244A CN 201811022344 A CN201811022344 A CN 201811022344A CN 109276244 A CN109276244 A CN 109276244A
- Authority
- CN
- China
- Prior art keywords
- signal
- brain
- value
- eeg signals
- neuron
- 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
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
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7235—Details of waveform analysis
- A61B5/7264—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems
Abstract
The recognition methods for age-care's intention based on brain wave information that the present invention provides a kind of, comprising: acquire the EEG signals under the movement of the elderly's difference and be converted into corresponding signal grayscale image；Signal grayscale image input convolutional layer is subjected to feature extraction；By obtained characteristic value input full articulamentum be fitted classification obtain the model of different EEG signals；Elderly brain electric signal is obtained to real-time monitoring to handle, and it is compared with model, judges the result of classification.
Description
Technical field
The present invention relates to a kind of EEG's Recognition sorting technique, especially a kind of the elderly's shield based on brain wave information
Manage the recognition methods being intended to.
Background technique
In today's society, age-care field still has following three main problem: the rhythm of 1. social life is not
Disconnected quickening makes the relatives for the personnel for needing to be nursed be difficult to extract time enough out；2. it is difficult that nursing staff recruits, keeping here is more difficult；And
And nursing staff's technical level is very different；3. corresponding nursing equipment both domestic and external requires to carry out relevant setting in advance, have
Certain use difficulty and obstacle.
EEG signals are a kind of apparent non-stationary signals, and since self-test measures EEG signals, people do it
Extensive work.As the acquisition research and analysis of EEG signals continues to develop, brain-computer interface becomes the novel man-machine friendship of people
Mutual mode has greatly reduced the people of existing the elderly and physical disabilities' care appliances so that human-computer interaction is simpler, direct
Border interacts difficulty；
Convolutional neural networks (Convolutional Neural Network, CNN) are developed recentlies, and are caused wide
A kind of efficient identification method of general attention.Its one kind can directly input the feedforward neural network of original image, research and development
But also new method and thinking occurs in signal analysis field, and it is widely used in area of pattern recognition.Its layout is closer
In actual biological neural network, explicit feature extraction is avoided, and is implicitly learnt from training data, weight is total
The complexity for reducing network is enjoyed, the complexity of data reconstruction in feature extraction and assorting process is avoided.
Summary of the invention
The recognition methods for age-care's intention based on brain wave information that the purpose of the present invention is to provide a kind of, passes through
EEG signals can accurately identify the desired movement of the elderly.
Realize the technical solution of the object of the invention are as follows: a kind of identification side that the age-care based on brain wave information is intended to
Method, comprising:
It acquires the EEG signals under the movement of the elderly's difference and is converted into corresponding signal grayscale image；
Signal grayscale image input convolutional layer is subjected to feature extraction；
By obtained characteristic value input full articulamentum be fitted classification obtain the model of different EEG signals；
Elderly brain electric signal is obtained to real-time monitoring to handle, and it is compared with model, judges classification
As a result.
The present invention feeds by imagining to user in brain, turns over, drains and drink water etc. the brain telecommunications of a series of actions
Number detection, and using obtain signal training one specific structure convolutional neural networks model to identify old personage, disability
The nursing of personage is intended to, to carry out brain signal classification, for the cranial nerve information of real-time monitoring the elderly and physical disabilities, at
It carries out after function identification to intellectual nursing bed control.
The invention will be further described with reference to the accompanying drawings of the specification.
Detailed description of the invention
Fig. 1 is the network architecture schematic diagram of convolutional layer.
Fig. 2 is the configuration diagram of full articulamentum.
Fig. 3 is network overall architecture schematic diagram.
Fig. 4 is the accuracy of model with the convergence curve schematic diagram of train epochs.
Fig. 5 is work flow diagram.
Specific embodiment
In conjunction with Fig. 5, a kind of recognition methods that the age-care based on brain wave information is intended to, concrete implementation scheme is such as
Under:
Step 1, the EEG signals for acquiring old personage or physical disabilities, are converted into corresponding signal grayscale image；And by its
It inputs convolutional layer and carries out feature extraction, obtained characteristic value is inputted into full articulamentum and is fitted classification
Step 2, the eeg data of real-time monitoring the elderly, and be inputted in model and compare classification；
Step 3, using the result of successful classification as intellectual nursing bed input signal；
Step 4, nursing bed receives signal, completes movement needed for it.
It is described that EEG signals are acquired to signal, it is the EEG signals using the production of Brain Products company of Germany
Acquisition equipment is acquired, and carries out signal using 32 channels.With the sample frequency of 16Hz, allow test subject imagine respectively feed,
It turns over, four kinds of movements of excretion and drinking-water, every kind of motion test 8s；
Every group of obtained signal is converted, specific conversion method is as follows:
Wherein, xi(t) signal, x are measured for i-th channel t momentmaxFor the maximum value of this group of signal, yiIt (t) is conversion
Afterwards, the signal of i-th of channel t moment.
After conversion, every group of signal can be converted to the grayscale image having a size of (32,16*8).
Obtained EEG signals gray scale pictures are subjected to input convolutional layer with 50 for batch.
The vector that the output feature flattening that convolutional layer is obtained is 2*56 at length, and it is input to full articulamentum.
Each middle layer uses Relu as activation primitive, and expression formula is as follows:
Y=max (0, x) (2)
The then expression formula of intrerneuron are as follows:
Its meaning be preceding layer the corresponding weight product of each neuron after sum, then using it as Relu function
Independent variable, output be next layer of neuron value.
In the training process, cross entropy loss function, expression formula have been used are as follows:
Wherein, what N was represented is the number of the neuron exported；y(i)Representative is the due output of i-th of neuron
Value；h(i)What is indicated is i-th of neuron real output value；What L was represented is cross entropy loss function value.
After obtaining cross entropy loss function, the instruction of network is carried out using the gradient decline optimizer in TensorFlow
Practice, so that loss function reaches minimum.The iterative formula of gradient descent algorithm are as follows:
Wherein, L is cross entropy loss function described above；It is obtained before kth walks iteration for i-th of weight
Weight；For i-th of weight kth walk iteration after obtained weight；α is learning rate, reflects each iteration weight variation
Degree size, be set as 0.000001 herein.Iterative steps stop operation when being 20000.
For output layer, using One-Hot coding form, i.e., in vector only one 1, remaining is all 0 coding form,
Available optimal effect.By brain the imagination feed output token be [1,0,0,0], brain the imagination turn over labeled as [0,
1,0,0], the excretion of the brain imagination is labeled as [0,0,1,0], and brain imagination drinking-water is labeled as [0,0,0,1], as every kind of data
Label uses the training for carrying out network model.
Classification and matching is carried out in the eeg data input model of acquisition, when similarity is more than certain threshold value, is then considered as point
Class success, shows that user has corresponding intention.
Embodiment
Brain neural signal data are obtained by equipment.Its frame number and port number should be identical as above-mentioned requirements, i.e. 32 channels,
The sample rate of 16Hz, the data of acquisition 8 seconds.Enough data should be acquired as far as possible.20 old will are employed herein
The data of hope person, each four groups of brains thinking nursing of every volunteer are intended to, and every group of movement is for 8 seconds, after acquisition, rest 20 seconds,
And next group of acquisition is carried out, 50 nursing of acquisition are intended to, and when volunteer experiences fatigue, can stop experiment and carry out adequately
Rest.
Collected electrical signal data format transformation at picture form, using 1 structure of network rack introduced in patent book,
75% is randomly selected in collected data, the training of parameter is carried out to network, obtains network model.
After training, using remaining 25% data as test set, it is input in trained model and is tested.
In training set, model has finally reached 67% accuracy, 64% accuracy has been reached in test set,
Significantly larger than 25% accuracy of random guess, it is believed that be effective model.
The test result for providing several groups test set data, thinks in testing herein, is as long as some element is greater than 0.7
It is regarded as 1；0 is regarded as less than 0.25；It is considered that an invalid number, the group are surveyed greater than 0.25 and less than 0.7
Test result is considered as mistake.
Find out from test result, 10 groups of test results, 6 groups are correctly, and the random guess also greater than 2.5 is correct
Number, it is believed that be effective
| There should be result [vector] | Actual result [vector] | Correctness |
| [1,0,0,0] [feed] | [0.835,0.127,0.182,0.110] | Correctly |
| [0,1,0,0] [turn] | [0.157,0.788,0.212,0.059] | Correctly |
| [0,0,1,0] [excretion] | [0.144,0.097,0.884,0.198] | Correctly |
| [0,0,0,1] [drinking-water] | [0.244,0.125,0.256,0.805] | Mistake |
| [1,0,0,0] [feed] | [0.695,0.128,0.219,0.116] | Mistake |
| [0,1,0,0] [turn] | [0.056,0.806,0.166,0.223] | Correctly |
| [0,0,1,0] [excretion] | [0.182,0.308,0.795,0.185] | Mistake |
| [0,0,0,1] [drinking-water] | [0.161,0.198,0.099,0.814] | Correctly |
| [1,0,0,0] [feed] | [0.721,0.158,0.214,0.132] | Mistake |
| [0,1,0,0] [turn] | [0.177,0.804,0.169,0.192] | Correctly |
Nursing bed is sent to induction signal by what successfully identification was intended to, it is controlled and completes respective action, that is, complete a work
Make circulation process, after return to work original state.
Claims (6)
1. a kind of recognition methods that the age-care based on brain wave information is intended to characterized by comprising
It acquires the EEG signals under the movement of the elderly's difference and is converted into corresponding signal grayscale image；
Signal grayscale image input convolutional layer is subjected to feature extraction；
By obtained characteristic value input full articulamentum be fitted classification obtain the model of different EEG signals；
Elderly brain electric signal is obtained to real-time monitoring to handle, and it is compared with model, judges the result of classification.
2. the method according to claim 1, wherein using the brain electricity of Brain Products company of Germany production
Signal collecting device is acquired, and acquires signal using 32 channels, and EEG signals are converted to corresponding signal ash by formula (1)
Degree figure
Wherein, xi(t) signal, x are measured for i-th channel t momentmaxFor the maximum value of this group of signal, yiIt (t) is after conversion the
The signal gray value of the channel i t moment；
Every group of signal can be converted to the grayscale image having a size of (32,16*8) after conversion.
3. according to the method described in claim 2, it is characterized in that, model acquisition methods include:
Obtained EEG signals gray scale pictures are subjected to input convolutional layer with 50 for batch；
The vector that the output feature flattening that convolutional layer is obtained is 2*56 at length, and it is input to full articulamentum；
Each middle layer of full articulamentum uses Relu as activation primitive, such as formula (2)
Y=max (0, x) (2)
Wherein, each neuron of preceding layerCorresponding weightAfter product sum after, by with as Relu function
Independent variable, output are the value of next layer of neuron, i.e. formula (3)
Wherein, full articulamentum input layer value is the vector that the output feature flattening that convolutional layer obtains is 2*56 at length；
Output layer is encoded using neuron value of the One-Hot coding form to the last layer.
4. according to the method described in claim 3, it is characterized in that,
Wherein, α is learning rate, and L is cross entropy loss function
y(i)For the due output valve of i-th of neuron, h(i)For i-th of neuron real output value.
5. according to the method described in claim 3, it is characterized in that, using One-Hot coding form coding result are as follows:
(1) output token of brain imagination feed is [1,0,0,0]；
(2) the brain imagination, which is turned over, is labeled as [0,1,0,0]；
(3) excretion of the brain imagination is labeled as [0,0,1,0]；
(4) brain imagination drinking-water is labeled as [0,0,0,1]；
Wherein, it is to think 1 that neuron value, which is greater than 0.7, and neuron value thinks 0 less than 0.25.
6. according to the method described in claim 5, it is characterized in that, feature by the EEG signals measured in real time through pulleying base
It extracts and the fitting of full articulamentum obtains real-time mark machine, real-time mark is compared with model, is judged old under the EEG signals
The movement of the imagination of year people.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811022344.7A CN109276244A (en) | 2018-09-03 | 2018-09-03 | The recognition methods that age-care based on brain wave information is intended to |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811022344.7A CN109276244A (en) | 2018-09-03 | 2018-09-03 | The recognition methods that age-care based on brain wave information is intended to |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109276244A true CN109276244A (en) | 2019-01-29 |
Family
ID=65184391
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811022344.7A Pending CN109276244A (en) | 2018-09-03 | 2018-09-03 | The recognition methods that age-care based on brain wave information is intended to |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109276244A (en) |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106821681A (en) * | 2017-02-27 | 2017-06-13 | 浙江工业大学 | A kind of upper limbs ectoskeleton control method and system based on Mental imagery |
| CN106909784A (en) * | 2017-02-24 | 2017-06-30 | 天津大学 | Epileptic electroencephalogram (eeg) recognition methods based on two-dimentional time-frequency image depth convolutional neural networks |
| CN106980367A (en) * | 2017-02-27 | 2017-07-25 | 浙江工业大学 | A kind of gesture identification method based on myoelectricity topographic map |
| CN107908928A (en) * | 2017-12-21 | 2018-04-13 | 天津科技大学 | A kind of hemoglobin Dynamic Spectrum Analysis Forecasting Methodology based on depth learning technology |
| CN107961007A (en) * | 2018-01-05 | 2018-04-27 | 重庆邮电大学 | A kind of electroencephalogramrecognition recognition method of combination convolutional neural networks and long memory network in short-term |
| CN108197648A (en) * | 2017-12-28 | 2018-06-22 | 华中科技大学 | A kind of Fault Diagnosis Method of Hydro-generating Unit and system based on LSTM deep learning models |
| CN108446020A (en) * | 2018-02-28 | 2018-08-24 | 天津大学 | Merge Mental imagery idea control method and the application of Visual Graph and deep learning |
-
2018
- 2018-09-03 CN CN201811022344.7A patent/CN109276244A/en active Pending
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106909784A (en) * | 2017-02-24 | 2017-06-30 | 天津大学 | Epileptic electroencephalogram (eeg) recognition methods based on two-dimentional time-frequency image depth convolutional neural networks |
| CN106821681A (en) * | 2017-02-27 | 2017-06-13 | 浙江工业大学 | A kind of upper limbs ectoskeleton control method and system based on Mental imagery |
| CN106980367A (en) * | 2017-02-27 | 2017-07-25 | 浙江工业大学 | A kind of gesture identification method based on myoelectricity topographic map |
| CN107908928A (en) * | 2017-12-21 | 2018-04-13 | 天津科技大学 | A kind of hemoglobin Dynamic Spectrum Analysis Forecasting Methodology based on depth learning technology |
| CN108197648A (en) * | 2017-12-28 | 2018-06-22 | 华中科技大学 | A kind of Fault Diagnosis Method of Hydro-generating Unit and system based on LSTM deep learning models |
| CN107961007A (en) * | 2018-01-05 | 2018-04-27 | 重庆邮电大学 | A kind of electroencephalogramrecognition recognition method of combination convolutional neural networks and long memory network in short-term |
| CN108446020A (en) * | 2018-02-28 | 2018-08-24 | 天津大学 | Merge Mental imagery idea control method and the application of Visual Graph and deep learning |
Non-Patent Citations (1)
| Title |
|---|
| 唐智川等: "基于深度卷积神经网络的运动想象分类及其在脑控外骨骼中的应用", 《计算机学报》 * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN102499677B (en) | Emotional state identification method based on electroencephalogram nonlinear features | |
| CN105956624B (en) | Mental imagery brain electricity classification method based on empty time-frequency optimization feature rarefaction representation | |
| CN108388890A (en) | A kind of neonatal pain degree assessment method and system based on human facial expression recognition | |
| CN111544856B (en) | Brain-myoelectricity intelligent full limb rehabilitation method based on novel transfer learning model | |
| CN105997064A (en) | Method for identifying human lower limb surface EMG signals (electromyographic signals) | |
| CN109255366B (en) | Emotional state adjusting system for online learning | |
| CN113288183A (en) | Silent voice recognition method based on facial neck surface myoelectricity | |
| CN106372729B (en) | Deep learning method and device for psychological analysis | |
| CN108814565A (en) | A kind of intelligent Chinese medicine health detection dressing table based on multi-sensor information fusion and deep learning | |
| CN111584029B (en) | Electroencephalogram self-adaptive model based on discriminant confrontation network and application of electroencephalogram self-adaptive model in rehabilitation | |
| CN109598222A (en) | Wavelet neural network Mental imagery brain electricity classification method based on the enhancing of EEMD data | |
| CN111493836B (en) | Postoperative acute pain prediction system based on brain-computer interface and deep learning and application | |
| CN106901941A (en) | A kind of joint of vertebral column depressurized system | |
| CN106056602A (en) | CNN (convolutional neural network)-based fMRI (functional magnetic resonance imaging) visual function data object extraction method | |
| CN111523617A (en) | Epilepsy detection system based on white matter fusion characteristic diagram and residual error attention network | |
| Borup et al. | Automatic sleep scoring using patient-specific ensemble models and knowledge distillation for ear-EEG data | |
| CN116595437B (en) | Training method, device and storage medium for zero calibration transfer learning classification model | |
| CN113171082A (en) | Method and device for evaluating injury of anterior cruciate ligament | |
| CN109498362A (en) | A kind of hemiplegic patient's hand movement function device for healing and training and model training method | |
| CN109276244A (en) | The recognition methods that age-care based on brain wave information is intended to | |
| Chueh et al. | Statistical prediction of emotional states by physiological signals with manova and machine learning | |
| Chen et al. | Trunk compensation electromyography features purification and classification model using generative adversarial network | |
| CN111967326A (en) | Gait recognition method based on lightweight multi-scale feature extraction | |
| CN116236209A (en) | Method for recognizing motor imagery electroencephalogram characteristics of dynamics change under single-side upper limb motion state | |
| Yao | Hearing loss classification via stationary wavelet entropy and cat swarm optimization |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190129 |
|
| RJ01 | Rejection of invention patent application after publication |