CN109886097A - A kind of method for detecting fatigue driving based on artificial fish school optimization H-ELM - Google Patents
A kind of method for detecting fatigue driving based on artificial fish school optimization H-ELM Download PDFInfo
- Publication number
- CN109886097A CN109886097A CN201910023261.8A CN201910023261A CN109886097A CN 109886097 A CN109886097 A CN 109886097A CN 201910023261 A CN201910023261 A CN 201910023261A CN 109886097 A CN109886097 A CN 109886097A
- Authority
- CN
- China
- Prior art keywords
- behavior
- hidden layer
- artificial fish
- feature
- elm
- 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
Landscapes
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention discloses a kind of method for detecting fatigue driving based on artificial fish school optimization H-ELM；Specifically: 1, lead using 32 brain wave acquisition equipment and obtain and drive EEG signals；2, original EEG signals are pre-processed, including frequency reducing, filtering；3, its power spectral density is obtained to pretreated progress Short Time Fourier Transform again, and frequency band division is carried out according to the frequency band of EEG signals, using the power of each frequency band as feature；4, to the feature of extraction using Multilayer Perception transfinite learning machine carry out classification learning, identification；5, the classification for the learning machine that transfinites, recognition effect are optimized by artificial fish-swarm algorithm.H-ELM classifier after the present invention is optimized using AFSA detects fatigue driving EEG signals, can effectively improve classification and Detection accuracy rate.
Description
Technical field
The present invention relates to the method for identifying and classifying of fatigue driving, in particular to a kind of to be based on artificial fish school optimization Multilayer Perception
Transfinite the method for detecting fatigue driving of learning machine.
Background technique
Before the learning machine (Extreme Learning Machine, ELM) that transfinites is a kind of novel rapid solving list hidden layer
Neural network method is presented, hidden layer deviation can be randomly generated in the training process, without adjustment, it is only necessary to which hidden layer mind is set
Number and activation primitive through member can obtain output weight by minimizing quadratic loss function.It therefore can be light
The good generalization ability of realization, Fast Learning ability and very strong approximation capability.Multilayer Perception transfinites learning machine (H-ELM)
Learn and indicate original input signal in the form of coding and decoding from coding function by constructing one, constantly reduces weight
Structure error achievees the purpose that characterize original signal, extracts better character representation, then pass through the common learning machine pair that transfinites
Feature is learnt and is classified.Since Multilayer Perception transfinites the characteristic of learning machine structure, the penalty factor of hidden layer and implicit
The number of plies of layer is set at random, and directly affects the classifying quality for the learning machine that transfinites.
Summary of the invention
The purpose of the present invention is using power spectrum to carry out feature extraction to original signal, in conjunction with Artificial Fish Swarm Optimization Algorithm
(AFSA) the hidden layer penalty factor of learning machine is transfinited to Multilayer Perception and the hidden layer number of plies is iterated optimizing, propose one kind
It is transfinited the driving fatigue detection method of learning machine (AFSA-H-ELM) based on artificial fish school optimization Multilayer Perception.
According to technical solution provided by the invention, proposing a kind of Multilayer Perception based on artificial fish school optimization transfinites study
Machine driving fatigue detection method, includes the following steps:
Step 1, the signal and non-fatigue driving state that fatigue driving state under 32 channels is obtained using brain wave acquisition equipment
Signal；
Step 2 carries out related pretreatment to original EEG signals, including frequency reducing, filtering；
Step 3, to pretreated carry out feature extraction: Fourier in short-term is carried out to treated EEG signals and is changed,
Then power spectral density is asked to the frequency band of treated EEG signals, EEG signals original in this way were converted to by power spectrum institute's generation
The feature vector of table；
Step 4 carries out classification learning, identification using the Multilayer Perception learning machine that transfinites to the feature of extraction；
Step 5, by artificial fish-swarm algorithm to Multilayer Perception transfinite learning machine hidden layer penalty factor and it is implicit layer by layer
Number two parameters of K optimize.
The frequency band of treated in the step 3 EEG signals is respectively δ (0.1-3Hz), θ (4-7Hz), α (8-
15Hz)、β(16-31Hz)、γ(32-50Hz)。
In the step 4, Multilayer Perception transfinite learning machine carry out classification learning, identification the step of specifically:
4-1. is by input signal Random Maps a to feature space；
4-2. passes through K layers of hidden layer, and each layer of hidden layer carries out unsupervised learning, the Η of outputKRepresent input data
High-level characteristic, feature is learnt and is classified by the common learning machine that transfinites again at this time；
Wherein the output of each layer of hidden layer is expressed as
Hi=g (Hi-1β),
Wherein, ΗiIt is the output of i-th of hidden layer, Ηi-1It is the output of (i-1)-th hidden layer, g () is hidden layer
Activation primitive, β are the output weights of hidden layer, from following formula calculating:
Wherein H=[h1,h2,...,hN] it is hidden layer output, X=[x1,x2,…,xN] it is input data.Punishing in norm
Penalty factor C is specified by user, is the tradeoff to frontier distance and training error is distinguished.
In the step 5, to the hidden layer penalty factor for the learning machine that transfinites and implied layer by layer by artificial fish-swarm algorithm
Number two parameters of K optimize, specific steps are as follows:
During initialization, parameter setting includes artificial fish-swarm scale fish_num, shoal of fish the number of iterations max_ to 5-1.
Gen, look for food maximum exploration number try_num, perceived distance Visual, moving step length Step and crowding factor delta.Initialize fish
Group's array is respectively two parameters of hidden layer penalty factor and hidden layer number of plies K in H-ELM；
5-2. first by input signal Random Maps to a feature space, then using K hidden layer to conversion after
Feature is learnt.The output Η of network layer implicit for this K layersKEffectively the advanced characterization of initial data has been arrived in study, most
Discriminance analysis is carried out to new feature according to classics ELM again afterwards；
5-3., which establishes Multilayer Perception using the parameter of initialization, to transfinite learning machine, is instructed according to training sample to the model
Practice, and calculates food concentration value, and take its maximum value；
Each shoal of fish of 5-4. simulates foraging behavior respectively, knock into the back behavior and behavior of bunching, and presses the maximum row of food concentration value
To execute, random behavior is executed if the behavior of missing；
Artificial fish-swarm algorithm includes foraging behavior, behavior of bunching, knock into the back behavior and random behavior；
1) current location of foraging behavior Artificial Fish individual is set as xi, the d within the scope of its present viewing fieldij≤ Visual, with
Machine selects a state xj, when maximizing, if f (xi) < f (xj), then xiBy formula (1) to xjShifting moves a step, otherwise selects again
State is selected, sees whether to meet advance condition, it is if being still unsatisfactory for condition, then random by formula (2) after attempting try_number times
Shifting moves a step；
xi+1=xi+rand×step (2)
Wherein di,j=| | xi-xj| | for the distance between Artificial Fish, Yi=f (xi) it is that i-th Artificial Fish is currently located position
The food concentration set, YiFor objective function, rand indicates a random number；
2) bunch behavior
Artificial Fish individual present position is xi, junior partner's number within the scope of its Visual is m, center xc,
If f (xc)/m > δ × f (xi), there is more food and less crowded at expression center, then xiAccording to formula (3) to xcMobile one
Step, if f (xc)/m < δ × f (xi), then implement foraging behavior, equally, if not seeing other partners, also executes foraging behavior；
3) it knocks into the back behavior
Artificial Fish individual present position is xi, food concentration value is maximum in all junior partners within the scope of its Visual
Position is xmax, number of partners m, if f (xmax)/m > δ × f (xi), show x in partnerjThere is more food and less gathers around
It squeezes, then xiBy formula (4) to xmaxShifting moves a step, if f (xmax)/m < δ × f (xi), then implement foraging behavior, equally, if do not had
See other junior partners, also executes foraging behavior；
4) random behavior be in a wider context search of food and companion, fish freely move about in water, Artificial Fish with
Machine behavior representation is arbitrarily mobile towards a direction in a certain position；I.e.
xi+1=xi+rand×Visual (5)
For 5-5. before being unsatisfactory for preset maximum number of iterations, artificial fish-swarm can continuously carry out various actions until meeting
Condition.After optimizing is completed, the position of the shoal of fish is hidden layer penalty factor and the hidden layer number of plies in this fatigue driving experiment
The optimal solution of two parameters of K；
5-6. learns the learning machine that transfinites with parameter building multilayer at this time, classifies to fatigue driving model.
The present invention has the beneficial effect that:
It, will be based on AFSA-H-ELM classification recognition result and H-ELM, ELM after carrying out feature extraction using power spectral density
Carry out Classification and Identification, traditional svm classifier recognition result compares, the results showed that, the H-ELM classification after being optimized using AFSA
Device accuracy in fatigue driving classification scene is higher, and effect is more preferable, effectively raises classification and Detection discrimination.
Detailed description of the invention
Fig. 1 is that Multilayer Perception transfinites learning machine schematic diagram；
Fig. 2 is AFSA-H-ELM iteration optimizing flow chart.
Specific embodiment:
The present invention is further explained in the light of specific embodiments.It is described below only as demonstration and explanation, not
It is intended that the present invention is limited in any way.
The step of present invention realizes as shown in Figures 1 and 2 is as follows:
Step 1 drives EEG signals using (the normal and fatigue) under brain wave acquisition equipment acquisition two states；
Step 2 carries out related pretreatment to original EEG signals, including frequency reducing, filtering；
Step 3, to pretreated carry out feature extraction: Fourier in short-term is carried out to treated EEG signals and is changed,
Then power spectral density is asked to the frequency band of treated EEG signals, EEG signals original in this way were converted to by power spectrum institute's generation
The feature vector of table；Wherein the frequency band of EEG signals is respectively δ (0.1-3Hz), θ (4-7Hz), α (8-15Hz), β (16-
31Hz),γ(32-50Hz)；At this point, the signal characteristic dimension of each channel has been reduced to five dimensions.
Step 4 carries out classification learning, identification using the Multilayer Perception learning machine that transfinites to the feature of extraction；
Specifically:
4-1. is by input signal Random Maps a to feature space；
4-2. passes through K layers of hidden layer, and each layer of hidden layer carries out unsupervised learning, the Η of outputKRepresent input data
High-level characteristic, feature is learnt and is classified by the common learning machine that transfinites again at this time；
Wherein the output of each layer of hidden layer is expressed as
Hi=g (Hi-1β),
Wherein, ΗiIt is the output of i-th of hidden layer, Ηi-1It is the output of (i-1)-th hidden layer, g () is hidden layer
Activation primitive, β are the output weights of hidden layer, from following formula calculating:
Wherein H=[h1,h2,...,hN] it is hidden layer output, X=[x1,x2,…,xN] it is input data.Punishing in norm
Penalty factor C is specified by user, is the tradeoff to frontier distance and training error is distinguished.
Step 5, by artificial fish-swarm algorithm to Multilayer Perception transfinite learning machine hidden layer penalty factor and it is implicit layer by layer
Number two parameters of K optimize.Specific steps are as follows:
During initialization, parameter setting includes artificial fish-swarm scale fish_num, shoal of fish the number of iterations max_ to 5-1.
Gen, look for food maximum exploration number try_num, perceived distance Visual, moving step length Step and crowding factor delta.Initialize fish
Group's array is respectively two parameters of hidden layer penalty factor and hidden layer number of plies K in H-ELM；
5-2. first by input signal Random Maps to a feature space, then using K hidden layer to conversion after
Feature is learnt.The output Η of network layer implicit for this K layersKEffectively the advanced characterization of initial data has been arrived in study, most
Discriminance analysis is carried out to new feature according to classics ELM again afterwards；
5-3., which establishes Multilayer Perception using the parameter of initialization, to transfinite learning machine, is instructed according to training sample to the model
Practice, and calculates food concentration value, and take its maximum value；
Each shoal of fish of 5-4. simulates foraging behavior respectively, knock into the back behavior and behavior of bunching, and presses the maximum row of food concentration value
To execute, random behavior is executed if the behavior of missing；
Artificial fish-swarm algorithm includes foraging behavior, behavior of bunching, knock into the back behavior and random behavior；
1) foraging behavior
The current location of Artificial Fish individual is set as xi, the d within the scope of its present viewing fieldij≤ Visual randomly chooses one
State xj, when maximizing, if f (xi) < f (xj), then xiBy formula (1) to xjShifting moves a step, otherwise reselects state, sees
The advance condition that whether meets is examined, after attempting try_number times, if being still unsatisfactory for condition, then presses one step of formula (2) random movement；
xi+1=xi+rand×step (2)
Wherein di,j=| | xi-xj| | for the distance between Artificial Fish, Yi=f (xi) it is that i-th Artificial Fish is currently located position
The food concentration set, YiFor objective function, rand indicates a random number；
2) bunch behavior
Artificial Fish individual present position is xi, junior partner's number within the scope of its Visual is m, center xc,
If f (xc)/m > δ × f (xi), there is more food and less crowded at expression center, then xiAccording to formula (3) to xcMobile one
Step, if f (xc)/m < δ × f (xi), then implement foraging behavior, equally, if not seeing other partners, also executes foraging behavior；
3) it knocks into the back behavior
Artificial Fish individual present position is xi, food concentration value is maximum in all junior partners within the scope of its Visual
Position is xmax, number of partners m, if f (xmax)/m > δ × f (xi), show x in partnerjThere is more food and less gathers around
It squeezes, then xiBy formula (4) to xmaxShifting moves a step, if f (xmax)/m < δ × f (xi), then implement foraging behavior, equally, if do not had
See other junior partners, also executes foraging behavior；
4) random behavior
For search of food in a wider context and companion, fish freely move about in water, and the random behavior of Artificial Fish indicates
It is arbitrarily mobile towards a direction in a certain position；I.e.
xi+1=xi+rand×Visual (5)
For 5-5. before being unsatisfactory for preset maximum number of iterations, artificial fish-swarm can continuously carry out various actions until meeting
Condition.After optimizing is completed, the position of the shoal of fish is hidden layer penalty factor and the hidden layer number of plies in this fatigue driving experiment
The optimal solution of two parameters of K；
5-6. learns the learning machine that transfinites with parameter building multilayer at this time, classifies to fatigue driving model.
Brain electricity sample when being driven using 1200 when 240 brain electricity samples are test data, uses respectively as training data
SVM, ELM, H-ELM and ASFA-H-ELM algorithm are classified, and classification results are as shown in table 1 below.
1 four kinds of sorting algorithm classification accuracy comparisons of table
| Sorting algorithm | Test accuracy rate |
| SVM | 83.47 |
| ELM | 77.29 |
| H-ELM | 87.00 |
| AFSA-H-ELM | 88.91 |
Can significantly find out from the contrast table of classification accuracy: compared to traditional SVM and ELM, the classification of H-ELM is quasi-
True rate is higher, is more suitable for the scene of fatigue driving experiment.And AFSA-H-ELM is advanced optimized on the basis of H-ELM algorithm
Parameter makes classification accuracy improve 1.9% or so, shows to effectively raise while AFSA-H-ELM obtains optimized parameter
Multilayer Perception transfinites the performance of learning machine.
Claims (4)
1. a kind of method for detecting fatigue driving based on artificial fish school optimization H-ELM, which is characterized in that this method specifically include as
Lower step:
Step 1 obtains the signal of fatigue driving state and the letter of non-fatigue driving state under 32 channels using brain wave acquisition equipment
Number；
Step 2 carries out related pretreatment to original EEG signals, including frequency reducing, filtering；
Step 3, to pretreated carry out feature extraction: Fourier in short-term is carried out to treated EEG signals and is changed, then
Power spectral density is asked to the frequency band of treated EEG signals, EEG signals original in this way are converted to as representated by power spectrum
Feature vector；
Step 4 carries out classification learning, identification using the Multilayer Perception learning machine that transfinites to the feature of extraction；
Step 5 transfinites the hidden layer penalty factor and hidden layer number of plies K of learning machine to Multilayer Perception by artificial fish-swarm algorithm
Two parameters optimize.
2. a kind of driving fatigue detection method based on artificial fish school optimization H-ELM according to claim 1, feature exist
In: the frequency band of treated in the step 3 EEG signals is respectively δ (0.1-3Hz), θ (4-7Hz), α (8-15Hz), β
(16-31Hz)、γ(32-50Hz)。
3. a kind of driving fatigue detection method based on artificial fish school optimization H-ELM according to claim 1, feature exist
In: in the step 4, Multilayer Perception transfinite learning machine carry out classification learning, identification the step of specifically:
4-1. is by input signal Random Maps a to feature space；
4-2. passes through K layers of hidden layer, and each layer of hidden layer carries out unsupervised learning, the Η of outputKRepresent the high level of input data
Feature is learnt and is classified to feature by the common learning machine that transfinites again at this time；
Wherein the output of each layer of hidden layer is expressed as
Hi=g (Hi-1β),
Wherein, ΗiIt is the output of i-th of hidden layer, Ηi-1It is the output of (i-1)-th hidden layer, g () is the activation of hidden layer
Function, β are the output weights of hidden layer, from following formula calculating:
Wherein H=[h1,h2,...,hN] it is hidden layer output, X=[x1,x2,…,xN] it is input data；l2Punishment in norm because
Sub- C is specified by user, is the tradeoff to frontier distance and training error is distinguished.
4. a kind of method for detecting fatigue driving based on artificial fish school optimization H-ELM according to claim 1, feature exist
In: in the step 5, by artificial fish-swarm algorithm to the hidden layer penalty factor and hidden layer number of plies K two of the learning machine that transfinites
A parameter optimizes, specific steps are as follows:
5-1. during initialization, parameter setting include artificial fish-swarm scale fish_num, shoal of fish the number of iterations max_gen,
Maximum of looking for food sounds out number try_num, perceived distance Visual, moving step length Step and crowding factor delta；Initialize shoal of fish number
Group is respectively two parameters of hidden layer penalty factor and hidden layer number of plies K in H-ELM；
5-2. is first by input signal Random Maps to a feature space, then using K hidden layer to the feature after conversion
Learnt；The output Η of network layer implicit for this K layersKEffectively the advanced characterization of initial data has been arrived in study, finally again
Discriminance analysis is carried out to new feature according to classics ELM；
5-3., which establishes Multilayer Perception using the parameter of initialization, to transfinite learning machine, is trained according to training sample to the model,
And food concentration value is calculated, and take its maximum value；
Each shoal of fish of 5-4. simulates foraging behavior respectively, knock into the back behavior and behavior of bunching, and holds by the maximum behavior of food concentration value
Row, executes random behavior if the behavior of missing；
Artificial fish-swarm algorithm includes foraging behavior, behavior of bunching, knock into the back behavior and random behavior；
1) foraging behavior
The current location of Artificial Fish individual is set as xi, the d within the scope of its present viewing fieldij≤ Visual randomly chooses a state
xj, when maximizing, if f (xi) < f (xj), then xiBy formula (1) to xjShifting moves a step, otherwise reselects state, and observation is
It is no to meet advance condition, after attempting try_number times, if being still unsatisfactory for condition, then press one step of formula (2) random movement；
xi+1=xi+rand×step (2)
Wherein di,j=| | xi-xj| | for the distance between Artificial Fish, Yi=f (xi) be i-th Artificial Fish present position food
Object concentration, YiFor objective function, rand indicates a random number；
2) bunch behavior
Artificial Fish individual present position is xi, junior partner's number within the scope of its Visual is m, center xcIf
f(xc)/m > δ × f (xi), there is more food and less crowded at expression center, then xiAccording to formula (3) to xcShifting moves a step, if f
(xc)/m < δ × f (xi), then implement foraging behavior, equally, if not seeing other partners, also executes foraging behavior；
3) it knocks into the back behavior
Artificial Fish individual present position is xi, the maximum position of food concentration value in all junior partners within the scope of its Visual
For xmax, number of partners m, if f (xmax)/m > δ × f (xi), show x in partnerjThere is more food and less crowded, then
xiBy formula (4) to xmaxShifting moves a step, if f (xmax)/m < δ × f (xi), then implement foraging behavior, equally, if not seeing it
He is junior partner, also executes foraging behavior；
4) random behavior
For search of food in a wider context and companion, fish freely move about in water, and the random behavior of Artificial Fish is expressed as
It is arbitrarily mobile towards a direction in a certain position；I.e.
xi+1=xi+rand×Visual (5)
For 5-5. before being unsatisfactory for preset maximum number of iterations, artificial fish-swarm can continuously carry out various actions until meeting item
Part；After optimizing is completed, the position of the shoal of fish is hidden layer penalty factor and hidden layer number of plies K in this fatigue driving experiment
The optimal solution of two parameters；
5-6. learns the learning machine that transfinites with parameter building multilayer at this time, classifies to fatigue driving model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910023261.8A CN109886097A (en) | 2019-01-10 | 2019-01-10 | A kind of method for detecting fatigue driving based on artificial fish school optimization H-ELM |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910023261.8A CN109886097A (en) | 2019-01-10 | 2019-01-10 | A kind of method for detecting fatigue driving based on artificial fish school optimization H-ELM |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109886097A true CN109886097A (en) | 2019-06-14 |
Family
ID=66925885
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910023261.8A Pending CN109886097A (en) | 2019-01-10 | 2019-01-10 | A kind of method for detecting fatigue driving based on artificial fish school optimization H-ELM |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109886097A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110477865A (en) * | 2019-08-14 | 2019-11-22 | 深圳先进技术研究院 | A kind of epileptic attack detection device, terminal device and storage medium |
| CN111783959A (en) * | 2020-07-08 | 2020-10-16 | 湖南工业大学 | Electronic skin touch pattern recognition method based on classification of hierarchical extreme learning machine |
| CN112348655A (en) * | 2020-09-29 | 2021-02-09 | 百维金科(上海)信息科技有限公司 | Credit evaluation method based on AFSA-ELM |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107451651A (en) * | 2017-07-28 | 2017-12-08 | 杭州电子科技大学 | A kind of driving fatigue detection method of the H ELM based on particle group optimizing |
| CN108229401A (en) * | 2018-01-05 | 2018-06-29 | 杭州电子科技大学 | A kind of multi-modal Modulation recognition method based on AFSA-SVM |
| CN108272463A (en) * | 2017-08-07 | 2018-07-13 | 上海交通大学 | A kind of flyer's fatigue detection method based on EEG signals |
-
2019
- 2019-01-10 CN CN201910023261.8A patent/CN109886097A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107451651A (en) * | 2017-07-28 | 2017-12-08 | 杭州电子科技大学 | A kind of driving fatigue detection method of the H ELM based on particle group optimizing |
| CN108272463A (en) * | 2017-08-07 | 2018-07-13 | 上海交通大学 | A kind of flyer's fatigue detection method based on EEG signals |
| CN108229401A (en) * | 2018-01-05 | 2018-06-29 | 杭州电子科技大学 | A kind of multi-modal Modulation recognition method based on AFSA-SVM |
Non-Patent Citations (4)
| Title |
|---|
| JIEXIONG TANG，ET AL.: "Extreme Learning Machine for Multilayer Perceptron", 《IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS》 * |
| TOBE: "贝叶斯优化：一种更好的参数调优方式", 《知乎：HTTPS://ZHUANLAN.ZHIHU.COM/P/29779000》 * |
| 张淞杰: "基于脑电信号识别的驾驶疲劳检测方法研究", 《万方数据》 * |
| 高志强，等.: "《深度学习 从入门到实战》", 30 June 2018, 中国铁道出版社 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110477865A (en) * | 2019-08-14 | 2019-11-22 | 深圳先进技术研究院 | A kind of epileptic attack detection device, terminal device and storage medium |
| CN111783959A (en) * | 2020-07-08 | 2020-10-16 | 湖南工业大学 | Electronic skin touch pattern recognition method based on classification of hierarchical extreme learning machine |
| CN112348655A (en) * | 2020-09-29 | 2021-02-09 | 百维金科(上海)信息科技有限公司 | Credit evaluation method based on AFSA-ELM |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106874956B (en) | The construction method of image classification convolutional neural networks structure | |
| CN104835507B (en) | A kind of fusion of multi-mode emotion information and recognition methods gone here and there and combined | |
| CN109784211A (en) | A kind of Mental imagery Method of EEG signals classification based on deep learning | |
| CN109886097A (en) | A kind of method for detecting fatigue driving based on artificial fish school optimization H-ELM | |
| CN112364779A (en) | Underwater sound target identification method based on signal processing and deep-shallow network multi-model fusion | |
| CN104035996B (en) | Field concept abstracting method based on Deep Learning | |
| CN109243494A (en) | Childhood emotional recognition methods based on the long memory network in short-term of multiple attention mechanism | |
| CN107085704A (en) | Fast face expression recognition method based on ELM own coding algorithms | |
| CN106779064A (en) | Deep neural network self-training method based on data characteristics | |
| CN108664950A (en) | A kind of electrical energy power quality disturbance identification and sorting technique based on deep learning | |
| CN106997475A (en) | A kind of insect image-recognizing method based on parallel-convolution neutral net | |
| CN110166154A (en) | A kind of software radio spectrum monitoring knowledge method for distinguishing neural network based | |
| CN108847279B (en) | Sleep breathing state automatic discrimination method and system based on pulse wave data | |
| CN110399846A (en) | A kind of gesture identification method based on multichannel electromyography signal correlation | |
| CN109389171A (en) | Medical image classification method based on more granularity convolution noise reduction autocoder technologies | |
| CN109512390B (en) | Sleep staging method and wearable device based on EEG time domain multi-dimensional features and M-WSVM | |
| CN109685071A (en) | Brain electricity classification method based on the study of common space pattern feature width | |
| CN107451651A (en) | A kind of driving fatigue detection method of the H ELM based on particle group optimizing | |
| CN109581339A (en) | A kind of sonar recognition methods based on brainstorming adjust automatically autoencoder network | |
| CN110287985B (en) | Depth neural network image identification method based on variable topology structure with variation particle swarm optimization | |
| CN115545086B (en) | Migratable feature automatic selection acoustic diagnosis method and system | |
| CN109740734A (en) | A kind of method of neuron spatial arrangement in optimization convolutional neural networks | |
| CN110096976A (en) | Human behavior micro-Doppler classification method based on sparse migration network | |
| CN104077598A (en) | Emotion recognition method based on speech fuzzy clustering | |
| Foong et al. | Convolutional neural network based rotten fruit detection using resnet50 |
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 |