CN106963374A - A kind of brain electro-detection method and device based on S-transformation and deep belief network - Google Patents

A kind of brain electro-detection method and device based on S-transformation and deep belief network Download PDF

Info

Publication number
CN106963374A
CN106963374A CN201710241861.2A CN201710241861A CN106963374A CN 106963374 A CN106963374 A CN 106963374A CN 201710241861 A CN201710241861 A CN 201710241861A CN 106963374 A CN106963374 A CN 106963374A
Authority
CN
China
Prior art keywords
brain
transformation
eeg signals
deep belief
section
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
Application number
CN201710241861.2A
Other languages
Chinese (zh)
Inventor
周卫东
陈丽艳
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Shandong University
Original Assignee
Shandong University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Shandong University filed Critical Shandong University
Priority to CN201710241861.2A priority Critical patent/CN106963374A/en
Publication of CN106963374A publication Critical patent/CN106963374A/en
Pending legal-status Critical Current

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
    • A61B5/24—Detecting, measuring or recording bioelectric or biomagnetic signals of the body or parts thereof
    • A61B5/316—Modalities, i.e. specific diagnostic methods
    • 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
    • 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 present invention relates to a kind of brain electro-detection method and device based on S-transformation and deep belief network, including：1) EEG signals are gathered, the EEG signals collected are subjected to A/D conversions, the EEG signals after storage A/D conversions；2) EEG signals after being changed to A/D carry out S-transformation, obtain brain electrical feature；3) to step 2) obtained brain electrical feature carries out linear normalization processing；4) by step 3) in the deep belief network of the brain electrical feature that extracts input, obtain principium identification result；5) principium identification result is subjected to glide filter, threshold decision, passage fusion, collar technical finesse by post-processing successively to principium identification result, obtains the normal or abnormal result of brain electricity, and be marked.The present invention extracts brain electrical feature with S-transformation, and carries out discriminant classification by deep belief network, not only clinically alleviates the workload of medical worker, and improve brain electro-detection efficiency and accuracy rate.

Description

A kind of brain electro-detection method and device based on S-transformation and deep belief network
Technical field
The present invention relates to a kind of utilization S-transformation and the brain electro-detection method and device of deep belief network, belong to brain electro-detection Technical field.
Technical background
Epilepsy is that the intermittent central nervous system function caused by a kind of repeatedly unexpected over-discharge with brain neuroblastoma member loses It is adjusted to the brain disorder of feature.So far, epilepsy detection is mainly entered by medical personnel by experience to electroencephalogram (EEG) Row estimates to complete, and checks that its workload is big, easily causes medical work whether containing characteristic waves such as epileptiform discharges in EEG Person is tired and produces erroneous judgement.Therefore, the automatic detection of electroencephalogram plays the role of important, and it can not only mitigate medical worker Workload, can also improve brain electro-detection accuracy and efficiency.
Since the sixties in last century, numerous scholars propose a variety of automatic seizure detection methods.At conventional brain electricity Reason method has Short Time Fourier Transform and wavelet transformation.Short Time Fourier Transform has some limitations, it is impossible to multiresolution Analyze the Time-Frequency Information of EEG signals.Though wavelet transformation realizes multiresolution analysis, its computation complexity is high.Conventional Brain electricity sorting algorithm has template matches, SVMs (SVM) etc..It is template that template matches, which choose typical epilepsy signal, will The waveform of detection carries out disease hair with template and recognized, but the diversity of EEG signals brings larger difficulty to the selection of sample.Support Vector machine is that supporting vector is solved by quadratic programming, solve quadratic programming and will be related to the calculating of high level matrix, and matrix is deposited Storage and calculating will expend substantial amounts of machine internal memory and operation time.
Chinese patent literature CN102429657A discloses a kind of epileptic EEG Signal classification and Detection method.This method is used The mode that wavelet analysis and approximate entropy are combined extracts brain electrical feature, and EEG signals are carried out with Neyman-Pearson criterions Classification.Though wavelet analysis realizes multiresolution analysis, its computation complexity is high, and analysis precision depends on the choosing of wavelet basis Select, with certain limitation.
The content of the invention
In view of the shortcomings of the prior art, the invention provides a kind of utilization S-transformation and the brain electro-detection side of deep belief network Method；
Device is realized present invention also offers above-mentioned brain electro-detection method；
What is proposed in the present invention has used the S-transformation of suitable treatment non-stationary signal, therefrom extracts power spectral density conduct Feature, saves the follow-up time with deep belief network (Deep Belief Network, DBN) training and classification, and keep away Exempt from the situation for feature over-fitting occur.
Summary of the invention：
The present invention extracts brain electrical feature with S-transformation first, and brain electrical feature then is carried out into linear normalization, and will be linear The deep belief network (DBN) of brain electrical feature feeding after normalization, obtains principium identification result, post-treated, including moving average Filtering, threshold decision, passage fusion, collar technology are final to obtain brain electro-detection result.
Term is explained：
1st, joint probability, refers to that in polynary probability distribution multiple stochastic variables meet the probability of respective condition respectively.
2nd, passage merge, in technical solution of the present invention, algorithm process be multiple passages eeg data, post processing In the stage, the result of multiple passages is integrated, if there are two or more passages to be judged in that is, multiple passages Breaking-out section is done, then this section of brain electricity is then judged to section of breaking out, otherwise is judged to do non-breaking-out section.
3rd, collar technology, in technical solution of the present invention, refers in post-processing stages, after passage fusion, by section of breaking out Each extension 0-2 sections of both sides.The starting of epileptic attack is all slow process, when patient just shows breaking-out illness, Brain electricity is also not up to threshold value after treatment, but has been breaking-out section, therefore with collar technology by this section of polishing.
The technical scheme is that：
A kind of brain electro-detection method of utilization S-transformation and deep belief network, including：
1) EEG signals are gathered, the EEG signals collected are subjected to A/D conversions, the EEG signals after storage A/D conversions；
2) to step 1) storage A/D conversion after EEG signals carry out S-transformation, obtain brain electrical feature；
3) to step 2) obtained brain electrical feature carries out linear normalization processing；
4) by step 3) in the deep belief network of brain electrical feature input after the linear normalization extracted processing, obtain preliminary Differentiate result；
5) by principium identification result through post processing, i.e., principium identification result is carried out successively glide filter, threshold decision, Passage fusion, collar technical finesse, obtain the normal or abnormal result of brain electricity, and be marked.
According to currently preferred, EEG signals are gathered by eeg amplifier and data collecting card；
It is further preferred that the model Neurofile NT of the eeg amplifier, the data collecting card is 16 A/D Change data capture card, sample frequency is 256Hz.
According to currently preferred, the step 2), the EEG signals after being changed to A/D carry out S-transformation, obtain brain electricity special Levy, including step is as follows：
(1) EEG signals after being changed to A/D are divided into mono- section of progress S-transformation of 2s, shown in S-transformation such as formula (I)：
In formula (I), (τ is f) S-transformation of EEG signals, τ is the position of Gaussian window on time shaft, and f is frequency, x (t) to S EEG signals after being changed for A/D, t is observation time；
After S-transformation, in time shaft, 2s is divided into 3 sections, every section of 0.67s is main due to EEG signals in frequency axis Frequency range is distributed in 1~30Hz, therefore 1~30Hz of interception, and according to 4 wave bands of EEG signals, 1~30Hz frequency ranges are divided into 1 This 4 parts of~4Hz, 4~8Hz, 8~12Hz, 12~30Hz, 4 wave bands include δ ripples, θ ripples, α ripples, β ripples, thus, every section of brain The S-transformation of electric signal is divided into 4*3=12 part in time-frequency domain；
(2) power spectral density is extracted in each part as brain electrical feature, as shown in formula (II), every section of EEG signals are carried Take out 12 brain electrical features：
Pi=E | S (τ, f) |2} (Ⅱ)
In formula (II), PiIt is brain electrical feature, i=1,2 ..., 12, E represent to be averaging.
According to currently preferred, the step 3), linear normalization processing is carried out to brain electrical feature by formula (III), thrown Shadow is between 0~1：
yi=(pi-MinValue)/(MaxValue-MinValue) (Ⅲ)
In formula (III), MinValue is brain electrical feature PiMinimum value, MaxValue be brain electrical feature PiMaximum, yi For the brain electrical feature after normalized, yi=y1,y2,...,yj,...,y12, y=(y1,y2,...,yj,...,y12)。
According to currently preferred, the step 4), by the feature y after the linear normalization extracted processing, input is deep Network output o is obtained in belief network (DBN)1And o2, principium identification result s=o is obtained after making the difference1-o2, the deep Belief Network Network includes 2 limitation Boltzmann machine RBM, i.e., including RBM1And RBM2, RBM1Including input layer, first hidden layer, RBM2Bag First hidden layer, second hidden layer are included, input layer (visual layers) neuron number is 12, first hidden layer neuron Number is 11, and second hidden layer neuron number is 5, and last layer of neuron number is 2, represents breaking-out and non-breaking-out two Class.
According to currently preferred, the step 5), including：
A, glide filter is carried out to principium identification result s, shown in glide filter such as formula (Ⅸ)：
In formula (Ⅸ), L is smooth window width, and z (k) is the value after principium identification result progress glide filter, s (k) the principium identification result of kth section EEG signals is referred to；
Decision content after b, the glide filter obtained to step a carries out threshold decision：It is labeled as during more than or equal to threshold value Th 0, represent non-breaking-out section；1 is labeled as during less than threshold value Th, breaking-out class is represented；
C, the obtained results of step b are entered row of channels fusion；For every section of eeg data, if in each and every one many passages, There are two or more passages to be determined as section of breaking out, then this segment data is determined as section of breaking out, is otherwise non-breaking-out section.
D, the result obtained to step c carry out collar technical finesse.It is the process of a gradual change because epilepsy signal breaks out, It occur frequently that signals are not up to situation of the threshold value with regard to completed stroke, therefore we need to solve this problem using collar technology, In the method, after threshold decision before and after section expand 0~2 section respectively.
According to currently preferred, smooth window width L=30.
The device that above-mentioned brain electro-detection method is realized, including eeg amplifier, data collecting card and the brain electricity being sequentially connected Detection module, using the eeg amplifier and data collecting card to being transferred to after eeg signal acquisition in brain electro-detection module, The power spectral density that every section of EEG signals are extracted by the use of the S-transformation module in brain electro-detection module uses linear normalizing as feature Change between feature normalization to 0~1, and feature is sent into degree of deeply convinceing network classifier module and classified, obtain tentatively Differentiate result, brain electro-detection result is obtained by post processing.
The present invention's has the advantages that：
The present invention extracts brain electrical feature using S-transformation, using deep belief network as grader, realizes that EEG signals are automatic Detection.The power spectral density of every segment signal is extracted with S-transformation first as feature, linear normalization then is carried out to feature； Then normalized feature is sent into deep belief network, finally gives the court verdict of EEG signals.Not only clinically mitigate The workload of medical worker, and improve detection efficiency and accuracy rate.
Brief description of the drawings
Fig. 1 is brain electro-detection method flow diagram of the present invention；
Fig. 2 realizes the hardware connection figure of the device of brain electro-detection method for the present invention；
Fig. 3 is present invention degree of deeply convinceing network of network structure chart；
Fig. 4 is last handling process schematic diagram of the present invention.
Embodiment
The present invention is further qualified with reference to Figure of description and embodiment, but not limited to this.
Embodiment 1
A kind of brain electro-detection method of utilization S-transformation and deep belief network, as shown in figure 1, including：
1) EEG signals are gathered by eeg amplifier and data collecting card, the EEG signals collected are carried out into A/D turns Change, the EEG signals after storage A/D conversions, the model Neurofile NT of eeg amplifier, data collecting card is that 16 A/D turn Data collecting card is changed, sample frequency is 256Hz；
2) to step 1) storage A/D conversion after EEG signals carry out S-transformation, obtain brain electrical feature；Including step such as Under：
(1) EEG signals after being changed to A/D are divided into mono- section of progress S-transformation of 2s, shown in S-transformation such as formula (I)：
In formula (I), (τ is f) S-transformation of EEG signals, τ is the position of Gaussian window on time shaft, and f is frequency, x (t) to S EEG signals after being changed for A/D, t is observation time；
After S-transformation, in time shaft, 2s is divided into 3 sections, every section of 0.67s is main due to EEG signals in frequency axis Frequency range is distributed in 1~30Hz, therefore 1~30Hz of interception, and according to 4 wave bands of EEG signals, 1~30Hz frequency ranges are divided into 1 This 4 parts of~4Hz, 4~8Hz, 8~12Hz, 12~30Hz, 4 wave bands include δ ripples, θ ripples, α ripples, β ripples, thus, every section of brain The S-transformation of electric signal is divided into 4*3=12 part in time-frequency domain；
(2) power spectral density is extracted in each part as brain electrical feature, as shown in formula (II), every section of EEG signals are carried Take out 12 brain electrical features：
Pi=E | S (τ, f) |2} (Ⅱ)
In formula (II), PiIt is brain electrical feature, i=1,2 ..., 12, E represent to be averaging.
3) to step 2) obtained brain electrical feature carries out linear normalization processing by formula (III) to brain electrical feature, projects to Between 0~1：
yi=(pi-MinValue)/(MaxValue-MinValue) (Ⅲ)
In formula (III), MinValue is brain electrical feature PiMinimum value, MaxValue be brain electrical feature PiMaximum, yi For the brain electrical feature after normalized, yi=y1,y2,...,yj,...,y12, y=(y1,y2,...,yj,...,y12)。
4) by step 3) feature y after the linear normalization extracted processing, net is obtained in the deep belief network (DBN) of input Network exports o1And o2, principium identification result s=o is obtained after making the difference1-o2, as shown in figure 3, deep belief network includes 2 limitation glass The graceful machine RBM of Wurz, i.e., including RBM1And RBM2, RBM1Including input layer, first hidden layer, RBM2Including first hidden layer, Second hidden layer, input layer (visual layers) neuron number is 12, and first hidden layer neuron number is 11, second Hidden layer neuron number is 5, and last layer of neuron number is 2, represents two classes of breaking-out and non-breaking-out.
5) by principium identification result through post processing, i.e., principium identification result is carried out successively glide filter, threshold decision, Passage fusion, collar technical finesse, obtain the normal or abnormal result of brain electricity, and be marked：As shown in figure 4, (a) is to deeply convince Spend network output preliminary judgement result, (b) be preliminary judgement result glide filter, (c) (d) (e) respectively represent passage 1~ Result of the 3 filtering outputs after threshold decision, (f) merges for 1~3 passage, and (g) is using the most termination after collar technology Really, wherein 0 represents non-breaking-out section, 1 represents breaking-out section.Including：
A, glide filter is carried out to principium identification result s, shown in glide filter such as formula (Ⅸ)：
In formula (Ⅸ), L is smooth window width, L=30, and z (k) is after principium identification result carries out glide filter Value, s (k) refers to the principium identification result of kth section EEG signals；
Decision content after b, the glide filter obtained to step a carries out threshold decision：It is labeled as during more than or equal to threshold value Th 0, represent non-breaking-out section；1 is labeled as during less than threshold value Th, breaking-out class is represented；
C, the obtained results of step b are entered row of channels fusion；For every section of eeg data, if in these three passages, There are two or more passages to be determined as section of breaking out, then this segment data is determined as section of breaking out, is otherwise non-breaking-out section.
D, the result obtained to step c carry out collar technical finesse.It is the process of a gradual change because epilepsy signal breaks out, It occur frequently that signals are not up to situation of the threshold value with regard to completed stroke, therefore we need to solve this problem using collar technology, In the method, after threshold decision before and after section expand 0~2 section respectively.
The brain electricity of 9 epileptics is detected using the method in the present embodiment, epileptic paradoxical discharge is detected Rate of accuracy reached 93.88%, false drop rate is 0.6/h.
Embodiment 2
Brain electro-detection method described in embodiment 1 realizes device, as shown in Fig. 2 electrically amplified including the brain being sequentially connected Device, data collecting card and brain electro-detection module, using the eeg amplifier and data collecting card to being passed after eeg signal acquisition In the defeated electro-detection module to brain, the power spectral density of every section of EEG signals is extracted using the S-transformation module in brain electro-detection module As feature, with linear normalization by between feature normalization to 0~1, and feature is sent into degree of deeply convinceing network classifier module It is middle to be classified, principium identification result is obtained, brain electro-detection result is obtained by post processing.

Claims (9)

1. a kind of brain electro-detection method of utilization S-transformation and deep belief network, it is characterised in that including：
1) EEG signals are gathered, the EEG signals collected are subjected to A/D conversions, the EEG signals after storage A/D conversions；
2) to step 1) storage A/D conversion after EEG signals carry out S-transformation, obtain brain electrical feature；
3) to step 2) obtained brain electrical feature carries out linear normalization processing；
4) by step 3) in the deep belief network of brain electrical feature input after the linear normalization extracted processing, obtain principium identification As a result；
5) principium identification result is subjected to glide filter, threshold decision, passage successively by post-processing to principium identification result Fusion, collar technical finesse, obtain the normal or abnormal result of brain electricity, and be marked.
2. the brain electro-detection method of a kind of utilization S-transformation according to claim 1 and deep belief network, it is characterised in that The step 2), the EEG signals after being changed to A/D carry out S-transformation, obtain brain electrical feature, including step is as follows：
(1) EEG signals after being changed to A/D are divided into mono- section of progress S-transformation of 2s, shown in S-transformation such as formula (I)：
S ( τ , f ) = ∫ - ∞ + ∞ x ( t ) | f | 2 π e - ( τ - t ) 2 f 2 2 e - i 2 π f t d t - - - ( I )
In formula (I), (τ, is f) S-transformation of EEG signals to S, and τ is the position of Gaussian window on time shaft, and f is frequency, and x (t) is A/D EEG signals after conversion, t is observation time；
After S-transformation, in time shaft, 2s is divided into 3 sections, every section of 0.67s, in frequency axis, intercepts 1~30Hz, according to brain telecommunications Number 4 wave bands, 1~30Hz frequency ranges are divided into 1~4Hz, 4~8Hz, 8~12Hz, this 4 parts of 12~30Hz, 4 ripples Section includes δ ripples, θ ripples, α ripples, β ripples, and thus, the S-transformation of every section of EEG signals is divided into 4*3=12 part in time-frequency domain；
(2) power spectral density is extracted in each part as brain electrical feature, as shown in formula (II), every section of EEG signals are extracted 12 brain electrical features：
Pi=E | S (τ, f) |2} (Ⅱ)
In formula (II), PiIt is brain electrical feature, i=1,2 ..., 12, E represent to be averaging.
3. the brain electro-detection method of a kind of utilization S-transformation according to claim 2 and deep belief network, it is characterised in that The step 3), linear normalization processing is carried out to brain electrical feature by formula (III), projected between 0~1：
yi=(pi-MinValue)/(MaxValue-MinValue) (Ⅲ)
In formula (III), MinValue is brain electrical feature PiMinimum value, MaxValue be brain electrical feature PiMaximum, yiTo return Brain electrical feature after one change processing, yi=y1,y2,...,yj,...,y12, y=(y1,y2,...,yj,...,y12)。
4. the brain electro-detection method of a kind of utilization S-transformation according to claim 3 and deep belief network, it is characterised in that The step 4), by the feature y after the linear normalization extracted processing, in the deep belief network of input, obtain network output o1 And o2, principium identification result s=o is obtained after making the difference1-o2, the deep belief network include 2 limitation Boltzmann machines, that is, include RBM1And RBM2, RBM1Including input layer, first hidden layer, RBM2Including first hidden layer, second hidden layer, input Layer neuron number is 12, and first hidden layer neuron number is 11, and second hidden layer neuron number is 5, finally One layer of neuron number is 2, represents two classes of breaking-out and non-breaking-out.
5. the brain electro-detection method of a kind of utilization S-transformation according to claim 4 and deep belief network, it is characterised in that The step 5), including：
A, glide filter is carried out to principium identification result s, shown in glide filter such as formula (Ⅸ)：
z ( k ) = 1 L Σ j = 1 - L 0 s ( k + j ) - - - ( I X )
In formula (Ⅸ), L is smooth window width, and z (k) is the value after principium identification result progress glide filter, and s (k) is Refer to the principium identification result of kth section EEG signals；
Decision content after b, the glide filter obtained to step a carries out threshold decision：0, table are labeled as during more than or equal to threshold value Th Show non-breaking-out section；1 is labeled as during less than threshold value Th, breaking-out class is represented；
C, the obtained results of step b are entered row of channels fusion；For every section of eeg data, if in each and every one many passages, there is two Individual or more than two passages are determined as section of breaking out, then this segment data is determined as section of breaking out, is otherwise non-breaking-out section；
D, the result obtained to step c carry out collar technical finesse.
6. the brain electro-detection method of a kind of utilization S-transformation according to claim 5 and deep belief network, it is characterised in that Smooth window width L=30.
7. the brain electro-detection method of a kind of utilization S-transformation according to claim 1 and deep belief network, it is characterised in that The step 1) EEG signals are gathered by eeg amplifier and data collecting card.
8. the brain electro-detection method of a kind of utilization S-transformation according to claim 7 and deep belief network, it is characterised in that The model Neurofile NT of described eeg amplifier, the data collecting card is 16 A/D change data capture cards, is adopted Sample frequency is 256Hz.
9. the device that any described brain electro-detection methods of claim 1-8 are realized, it is characterised in that including the brain being sequentially connected EEG signals are adopted by electric amplifier, data collecting card and brain electro-detection module using the eeg amplifier and data collecting card It is transferred to after collection in brain electro-detection module, the power of every section of EEG signals is extracted using the S-transformation module in brain electro-detection module Feature with linear normalization by between feature normalization to 0~1, and is sent into degree of deeply convinceing network class as feature by spectrum density Classified in device module, obtain principium identification result, brain electro-detection result is obtained by post processing.
CN201710241861.2A 2017-04-14 2017-04-14 A kind of brain electro-detection method and device based on S-transformation and deep belief network Pending CN106963374A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201710241861.2A CN106963374A (en) 2017-04-14 2017-04-14 A kind of brain electro-detection method and device based on S-transformation and deep belief network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201710241861.2A CN106963374A (en) 2017-04-14 2017-04-14 A kind of brain electro-detection method and device based on S-transformation and deep belief network

Publications (1)

Publication Number Publication Date
CN106963374A true CN106963374A (en) 2017-07-21

Family

ID=59332583

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201710241861.2A Pending CN106963374A (en) 2017-04-14 2017-04-14 A kind of brain electro-detection method and device based on S-transformation and deep belief network

Country Status (1)

Country Link
CN (1) CN106963374A (en)

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107587955A (en) * 2017-07-28 2018-01-16 大连理工大学 The scaling method of rocket engine thrust offset based on deep belief network
CN107714035A (en) * 2017-09-14 2018-02-23 中国人民解放军第四军医大学 A kind of wearable digitlization eeg monitoring helmet
CN112116995A (en) * 2020-08-31 2020-12-22 山东师范大学 Brain U nursing machine and method
CN113057653A (en) * 2021-03-19 2021-07-02 浙江科技学院 Channel mixed convolution neural network-based motor electroencephalogram signal classification method

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104166548A (en) * 2014-08-08 2014-11-26 同济大学 Deep learning method based on motor imagery electroencephalogram data
CN104749432A (en) * 2015-03-12 2015-07-01 西安电子科技大学 Estimation method of multi-component non-stationary signal instantaneous frequency based on focusing S-transform
CN106264519A (en) * 2015-06-01 2017-01-04 山东大学苏州研究院 A kind of epileptic electroencephalogram (eeg) detection device based on Stockwell conversion

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104166548A (en) * 2014-08-08 2014-11-26 同济大学 Deep learning method based on motor imagery electroencephalogram data
CN104749432A (en) * 2015-03-12 2015-07-01 西安电子科技大学 Estimation method of multi-component non-stationary signal instantaneous frequency based on focusing S-transform
CN106264519A (en) * 2015-06-01 2017-01-04 山东大学苏州研究院 A kind of epileptic electroencephalogram (eeg) detection device based on Stockwell conversion

Non-Patent Citations (5)

* Cited by examiner, † Cited by third party
Title
DRAUSIN WULSIN等: "Semi-Supervised Anomaly Detection for EEG Waveforms Using Deep Belief Nets", 《2010 NINTH INTERNATIONAL CONFERENCE ON MACHINE LEARNING AND APPLICATIONS》 *
PINYI LI等: "Single-channel EEG-based mental fatigue detection based on deep belief network", 《2016 38TH ANNUAL INTERNATIONAL CONFERENCE OF THE IEEE ENGINEERING IN MEDICINE AND BIOLOGY SOCIETY (EMBC)》 *
严爱玉: "基于Stockwell变换与Boosting算法的自动癫痫检测", 《中国优秀硕士学位论文全文数据库医药卫生科技辑》 *
夏玉丹: "基于s变换和奇异值分解的自动癫痫检测算法", 《中国优秀硕士学位论文全文数据库医药卫生科技辑》 *
徐朝阳: "基于深度网络的脑电信号分类方法研究", 《中国优秀硕士学位论文全文数据库医药卫生科技辑》 *

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107587955A (en) * 2017-07-28 2018-01-16 大连理工大学 The scaling method of rocket engine thrust offset based on deep belief network
CN107587955B (en) * 2017-07-28 2019-03-05 大连理工大学 The scaling method of rocket engine thrust offset based on deep belief network
CN107714035A (en) * 2017-09-14 2018-02-23 中国人民解放军第四军医大学 A kind of wearable digitlization eeg monitoring helmet
CN112116995A (en) * 2020-08-31 2020-12-22 山东师范大学 Brain U nursing machine and method
CN113057653A (en) * 2021-03-19 2021-07-02 浙江科技学院 Channel mixed convolution neural network-based motor electroencephalogram signal classification method

Similar Documents

Publication Publication Date Title
CN110811609B (en) Epileptic spike intelligent detection device based on self-adaptive template matching and machine learning algorithm fusion
CN104586387B (en) Method for extracting and fusing time, frequency and space domain multi-parameter electroencephalogram characters
Kaya et al. 1D-local binary pattern based feature extraction for classification of epileptic EEG signals
CN110432898A (en) A kind of epileptic attack eeg signal classification system based on Nonlinear Dynamical Characteristics
CN111329474B (en) Electroencephalogram identity recognition method and system based on deep learning and information updating method
CN106963374A (en) A kind of brain electro-detection method and device based on S-transformation and deep belief network
CN106803081A (en) A kind of brain electricity sorting technique based on Multi-classifers integrated
CN106473750A (en) Personal identification method based on photoplethysmographic optimal period waveform
CN110321783A (en) A kind of MEG spike detection method and system based on 1D convolutional neural networks
CN103190904B (en) Electroencephalogram classification detection device based on lacuna characteristics
CN104809434A (en) Sleep staging method based on single-channel electroencephalogram signal ocular artifact removal
Li et al. Analysis and classification of EEG signals using a hybrid clustering technique
CN105942974A (en) Sleep analysis method and system based on low frequency electroencephalogram
CN106388814A (en) Epilepsy electroencephalogram signal identification method based on optimal kernel time-frequency distribution visibility graph
Du et al. A method for detecting high-frequency oscillations using semi-supervised k-means and mean shift clustering
Prasanth et al. Deep learning for interictal epileptiform spike detection from scalp EEG frequency sub bands
CN114266276A (en) Motor imagery electroencephalogram signal classification method based on channel attention and multi-scale time domain convolution
CN111144269B (en) Signal correlation behavior recognition method and system based on deep learning
CN114081505A (en) Electroencephalogram signal identification method based on Pearson correlation coefficient and convolutional neural network
CN110464371A (en) Method for detecting fatigue driving and system based on machine learning
Ebrahimpour et al. Epileptic seizure detection using a neural network ensemble method and wavelet transform
Zan et al. Local Pattern Transformation-Based convolutional neural network for sleep stage scoring
Gnana Rajesh Analysis of MFCC features for EEG signal classification
CN109144277B (en) Method for constructing intelligent vehicle controlled by brain based on machine learning
Shankar et al. Discrimination of types of seizure using brain rhythms based on markov transition field and deep learning

Legal Events

Date Code Title Description
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: 20170721

RJ01 Rejection of invention patent application after publication