CN111950366B - Convolutional neural network motor imagery electroencephalogram classification method based on data enhancement - Google Patents
Convolutional neural network motor imagery electroencephalogram classification method based on data enhancement Download PDFInfo
- Publication number
- CN111950366B CN111950366B CN202010650831.9A CN202010650831A CN111950366B CN 111950366 B CN111950366 B CN 111950366B CN 202010650831 A CN202010650831 A CN 202010650831A CN 111950366 B CN111950366 B CN 111950366B
- Authority
- CN
- China
- Prior art keywords
- data
- electroencephalogram
- layer
- neural network
- convolutional neural
- 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.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 42
- 238000013527 convolutional neural network Methods 0.000 title claims abstract description 30
- 238000013526 transfer learning Methods 0.000 claims abstract description 7
- 238000012549 training Methods 0.000 claims description 20
- 210000002569 neuron Anatomy 0.000 claims description 15
- 238000011176 pooling Methods 0.000 claims description 7
- 238000012360 testing method Methods 0.000 claims description 7
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 claims description 3
- 230000004913 activation Effects 0.000 claims description 3
- 230000006870 function Effects 0.000 claims description 3
- 230000008569 process Effects 0.000 claims description 3
- 238000002790 cross-validation Methods 0.000 abstract description 6
- 210000004556 brain Anatomy 0.000 description 6
- 230000033001 locomotion Effects 0.000 description 5
- 238000010586 diagram Methods 0.000 description 4
- 238000001914 filtration Methods 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 3
- 238000002474 experimental method Methods 0.000 description 3
- 238000000605 extraction Methods 0.000 description 3
- 210000005036 nerve Anatomy 0.000 description 3
- 238000013136 deep learning model Methods 0.000 description 2
- 230000003447 ipsilateral effect Effects 0.000 description 2
- 210000003205 muscle Anatomy 0.000 description 2
- 238000003909 pattern recognition Methods 0.000 description 2
- 230000008439 repair process Effects 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 206010008190 Cerebrovascular accident Diseases 0.000 description 1
- 206010019468 Hemiplegia Diseases 0.000 description 1
- 208000016285 Movement disease Diseases 0.000 description 1
- 208000006011 Stroke Diseases 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 230000007177 brain activity Effects 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 238000013480 data collection Methods 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 230000005012 migration Effects 0.000 description 1
- 238000013508 migration Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 210000000578 peripheral nerve Anatomy 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 230000002269 spontaneous effect Effects 0.000 description 1
- 230000003797 telogen phase Effects 0.000 description 1
- 230000002123 temporal effect Effects 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
- 238000000844 transformation 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
- G06F2218/12—Classification; Matching
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F3/00—Input arrangements for transferring data to be processed into a form capable of being handled by the computer; Output arrangements for transferring data from processing unit to output unit, e.g. interface arrangements
- G06F3/01—Input arrangements or combined input and output arrangements for interaction between user and computer
- G06F3/011—Arrangements for interaction with the human body, e.g. for user immersion in virtual reality
- G06F3/015—Input arrangements based on nervous system activity detection, e.g. brain waves [EEG] detection, electromyograms [EMG] detection, electrodermal response detection
-
- 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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2203/00—Indexing scheme relating to G06F3/00 - G06F3/048
- G06F2203/01—Indexing scheme relating to G06F3/01
- G06F2203/011—Emotion or mood input determined on the basis of sensed human body parameters such as pulse, heart rate or beat, temperature of skin, facial expressions, iris, voice pitch, brain activity patterns
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Computing Systems (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- Data Mining & Analysis (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Mathematical Physics (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Signal Processing (AREA)
- Dermatology (AREA)
- Neurology (AREA)
- Neurosurgery (AREA)
- Human Computer Interaction (AREA)
- Measurement And Recording Of Electrical Phenomena And Electrical Characteristics Of The Living Body (AREA)
Abstract
The invention discloses a motor imagery electroencephalogram signal classification method based on a data enhancement convolutional neural network, which carries out five-time cross validation on 64 channel signals of all subjects to obtain the average accuracy of two, three and four classification tasks of 87.32%,76.26% and 64.72% respectively. The global classifier is applied to the classification of the electroencephalogram signals of the single subject by utilizing the transfer learning, and the average accuracy rate of the global classifier reaches 91.06%,82.76% and 73.46%; compared with the Dose work, the architecture provided by the invention has better performance; after data enhancement, data enhancement is performed on the four classes of classified data. The average accuracy rate is respectively improved from 64.72% to 66.73% (global model), and 73.46% to 76.78% (subject model), which shows that the data enhancement method can effectively improve the accuracy rate of electroencephalogram signal classification.
Description
Technical Field
The invention relates to the field of brain-computer interfaces and pattern recognition, in particular to a motor imagery electroencephalogram signal classification method based on a data enhancement convolutional neural network.
Background
Brain-computer interface (Brain Computer Interface, BCI) technology is an emerging technology, and has received a great deal of attention in recent years. BCI creates a new non-muscle interaction channel that humans can interact with the surrounding environment using signals generated by brain activity without the involvement of the peripheral nerves and muscles. By pattern recognition of the signals, communication of the intent of the person to external devices such as computers, speech synthesizers, auxiliary devices and nerve repair is achieved. The brain-computer interface has important application in the medical field, the neurobiology field and the psychological field. An Electroencephalogram (EEG) based on Motor Imagery (MI) is an endogenous spontaneous Electroencephalogram, has the advantages of being simple to operate, flexible in paradigm, low in cost, good in portability, low in risk and the like, has been widely studied, and has wide application potential in the fields of nerve rehabilitation, nerve repair, games and the like. The rehabilitation training device not only can help patients with serious movement disorders such as apoplexy hemiplegia to carry out rehabilitation training, and control equipment can realize self-care, but also can entertain the general masses, such as various brain games combined with virtual reality technology.
The brain-computer interface system generally comprises a signal acquisition part, a preprocessing or signal enhancement part, a characteristic extraction part, a classification and discrimination part and a control interface part. Improving the accuracy of electroencephalogram classification is the core of the whole system, namely an effective signal feature extraction and classification method to convert the electroencephalogram of a motion image into control of equipment or an application program. Because the signals have the characteristics of large individual difference, low signal-to-noise ratio, unstable signals and the like, the classification performance of the signals and the accuracy of model training are affected. At present, the electroencephalogram data based on motor imagery is scarce, the electroencephalogram data is relatively expensive to collect, the privacy safety problem is also involved, and the electroencephalogram data based on motor imagery has fewer public data sets and fewer sample numbers.
Disclosure of Invention
Accordingly, it is an object of the present invention to provide a model training method based on convolutional neural networks (Convolutional Neural Networks, CNN); on the other hand, the data enhancement method is further provided on the basis of the convolutional neural network model, and by the method, the deficiency of the brain electrical data based on motor imagery can be compensated, and the classification accuracy of signals is improved.
A motor imagery electroencephalogram signal classification method comprises the following steps:
acquiring electroencephalogram signal data of at least two types of motor imagery tasks in a Physionet data set;
inputting the electroencephalogram signal data into a convolutional neural network for training, thereby realizing classification of the electroencephalogram signal data;
the first layer of the convolutional neural network is a convolutional layer, convolutional operation is carried out on the electroencephalogram signal data along a time axis, and the size of output data is consistent with that of input data;
the second layer is a convolution layer, convolves the EEG signal along the EEG channel axis, and reduces the output size to half of the input;
the third layer is a maximum pooling layer for pooling the EEG signals along a time axis, the size of a kernel is 30 samples, and the step length is 15; flattening the layered structure to form 6300 single-dimensional neurons;
the last three layers are three fully connected layers, the first fully connected layer reduces the number of neurons from 6300 to 100, the second fully connected layer reduces the number of neurons from 100 to 32, the last fully connected layer is a softmax layer, and the number of neurons is reduced from 32 to the number of neurons in the data to be classified.
Further, the electroencephalogram signal data is subjected to data enhancement and then training, specifically:
normalizing the electroencephalogram data of each channel of each test for the original electroencephalogram data in the Physionet data set;
and randomly selecting more than two groups of electroencephalogram signal data of the same type of labels, superposing and standardizing the electroencephalogram signal data according to the channels, and generating new electroencephalogram signal data.
Preferably, the first layer of the convolutional neural network employs 100 filters, the kernel size is 30 samples, and the step size is 1.
Preferably, the same padding method is adopted to fill 0 in the electroencephalogram signal data so that the output size is the same as the input size after convolution operation.
Preferably, a convolutional neural network uses 100 filters, the kernel size is 30 samples, and the step size is 1.
Preferably, the third layer of the convolutional neural network has a kernel size of 30 samples and a step size of 15.
Preferably, the convolutional neural network employs Adam optimizers, reLU activation and a cost function of class cross entropy.
The invention has the following beneficial effects:
according to the motor imagery electroencephalogram signal classification method based on the data enhancement convolutional neural network, the 64-channel signals of all subjects are subjected to five-time cross validation to obtain the average accuracy of two, three and four classification tasks which respectively reach 87.32%,76.26% and 64.72%. The global classifier is applied to the classification of the brain electrical signals of the single subject by using transfer learning, and the average accuracy rate of the global classifier reaches 91.06%,82.76% and 73.46%. Compared with the Dose work, the architecture provided by the invention has better performance.
After data enhancement, data enhancement is performed on the four classes of classified data. The average accuracy rate is respectively improved from 64.72% to 66.73% (global model), and 73.46% to 76.78% (subject model), which shows that the data enhancement method can effectively improve the accuracy rate of electroencephalogram signal classification.
Drawings
FIG. 1 is a block diagram of a training process of the present invention;
FIG. 2 is an average accuracy rate after data enhancement in global mode;
FIG. 3 is the average accuracy after data enhancement in subject mode;
FIG. 4 shows the average accuracy of two training models before and after data enhancement;
FIG. 5 is a schematic diagram of a convolutional neural network framework;
fig. 6 is a data enhancement schematic.
Detailed Description
The invention will now be described in detail by way of example with reference to the accompanying drawings.
The invention is divided into two parts, wherein the first part is the research of a convolutional neural network structure, and the second part is the research of a data enhancement method based on the network structure.
The whole experimental procedure is shown in figure 1. The invention adopts a public data set: physionet data set (brain electrical signals are acquired when a subject imagines movement of any part of the body). In the model training process, training is carried out by adopting a convolutional neural network method, and then a new data enhancement method is further provided for training on the basis of the convolutional neural network. Five-fold cross-validation methods are employed, and the above models are collectively referred to as global patterns. In addition, a migration learning method is introduced into the two training models, the global classifier is applied to the classification of the electroencephalogram signals of the single subject, and a quadruple cross-validation method is adopted. I.e., on the basis of a global classifier, training on 3/4 of the data of a single subject, the remaining 1/4 is tested. The model after the transfer learning is collectively referred to as a subject model.
1. Data set:
the current electroencephalogram data based on motor imagery is scarce, the electroencephalogram data collection is relatively expensive, meanwhile, the privacy safety problem is also related, and the public data set is few. The invention uses a public data set: physionet data set consisting of more than 1500 electroencephalograms for one and two minutes, a total of 109 subjects engaged in the experiment. Each subject performed 14 large sets of experiments, including two parts, a motor executive task and a motor imagery task, and 64 channels of electroencephalogram data were recorded using the BCI2000 system. In this embodiment, only data of motor imagery tasks are used, including 4 imagery motions (opening and closing of left palm, opening and closing of right palm, opening and closing of both palms, opening and closing of both feet) and a baseline task (eye opening or eye closing rest phase).
Data classification cases are as shown for two classifications: the dataset contained experimental data for right and left hand opening and closing, 21 sets of experimental data per class, 42 sets of data per subject. For three classifications: and adding 21 groups of rest state task data into the two kinds of data, namely, the subjects do not execute any motor imagery tasks under the condition of opening eyes or closing eyes, and each subject has 63 groups of data. For four classifications: to the three-class data, 21 sets of data for both foot opening and closing, i.e., 84 sets of data for each subject were added. Each set of data contains a data length of 6 seconds (960 samples), a motor imagery of 4 seconds and an idle time of 1 second before and after.
2. Convolutional neural network structure:
the structural schematic diagram of the convolutional neural network is shown in fig. 5, and the structural schematic diagram from left to right is shown as follows: C100@30X1-C100@1X10X 30-P@30X1-Flatten-FC@100-FC@32-Softmax.
The first layer performs convolution on the electroencephalogram signal along a time axis. 100 filters are used, the kernel size is 30 samples, the step size is 1, and the same padding method is adopted, namely 0 is filled around the input data, so that the output size is the same as the input after the convolution operation, and the dimension of the architecture is reserved. This layer corresponds to linear pre-filtering of the electroencephalogram signal for each channel, known as temporal filtering.
The second layer is convolved along the EEG channel axis of the EEG signal. 100 filters are used, the kernel size is 30 samples, the step size is 1, the valid padding method is adopted, namely, the input is not filled around, the output size can be reduced after convolution operation, and the channel size is halved, so that the influence of an inefficient channel is reduced. This layer can reduce the dimensionality between channels and the impact of motion-independent regions, known as spatial filtering (spatial filtering).
The third layer is the maximum pooling layer (Maxpoolinglayer). Pooling was performed along the time axis, with a kernel size of 30 samples and a step size of 15. The use of the largest pool layer to perform dimension reduction has some characteristics in the feature map that can be well summarized and provides a more robust architecture.
Next, the structure after the maximum pooling layer is flattened (flat) to form 6300 single-dimensional neurons. Followed by three Fully Connected (FC) layers. The first fully connected layer reduces the number of neurons from 6300 to 100, the second fully connected layer reduces the number of neurons from 100 to 32, the last fully connected layer, also called softmax layer, reduces the number of neurons from 32 to the number of neurons in the data to be classified (n-class).
In this architecture, the Adam optimizer, reLU activation, and cost function of class cross entropy are used.
The model obtained by carrying out five-fold cross validation on 64-channel signals of all subjects based on the convolutional neural network structure is called a global model (Global model), and the average accuracy of two, three and four classification tasks respectively reaches 87.32%,76.26% and 64.72%. The global classifier is applied to the electroencephalogram signal classification of a single subject by using transfer learning, and the model obtained after the transfer learning is called a subject model (subject), and the average cross-validation accuracy of the model reaches 91.06%,82.76% and 73.46%. After removing the data set of BCI illiterates (subjects with classification accuracy lower than the blind guess), the average accuracy of the 104 tested two, three, and four classification tasks in the global model was 88.32%, 78.02%, 66.36%, respectively. The average accuracy of the classification tasks of the second class, the third class and the fourth class based on the topic model is 91.92 percent, 83.78 percent and 74.60 percent respectively. We propose an architecture that performs better than the task of Dose.
Only 16 channel data related to the motion area are selected, and data processing is carried out by adopting the same method, so that the average accuracy of classification tasks of two, three and four types in the global mode is 83.57%,72.23% and 59.74% respectively. The average accuracy of the classification tasks of the second class, the third class and the fourth class in the subject mode is 87.88%,79.64% and 68.03% respectively. As shown in table 1, the neural network structure performance of the present invention is more excellent compared with that of other documents.
Table 1: the left/right hand motor imagery on the Physionet dataset was studied differently for the classification results. Including the number of electroencephalogram channels used, the training mode (global model or subject model), the average accuracy achieved, and the methods used for feature extraction and classification.
3. Data enhancement:
training an effective deep learning model requires a large amount of training data because a large amount of parameters in the deep learning model need to be adjusted, and the data enhancement technique can well solve the problem of insufficient data. In many fields, new data samples are artificially generated by shifting, scaling, rotating, etc., existing training data. However, because of the non-stationarity of the electroencephalogram signals, these similar geometric transformations are not applicable to electroencephalogram signals. Therefore, it is important to select a suitable data enhancement method, so that the characteristics of the original signal can be maintained.
The results of Pfurtscheller et al show that right hand motor imagery is accompanied by contralateral beta event-related walkout and ipsilateral beta event-related synchronization. Based on the above, we invented a new data expansion method, namely a superposition method, the main idea is to superpose and normalize motor imagery electroencephalogram data with the same type of labels to generate new artificial data. The method comprises the following specific steps:
extracting original electroencephalogram data based on motor imagery of 64 electroencephalogram channels of 109 subjects and corresponding labels thereof according to the requirements of the first partial data set;
the electroencephalogram data for each channel of each trial was normalized. (the original electroencephalogram signal value of the current test of a certain channel is subtracted from the average value of the original electroencephalogram signal of the current test of the channel and then divided by the standard deviation of the original electroencephalogram signal of the current test of the channel).
And randomly selecting n (n is a parameter) similar labels for test, and superposing and standardizing n groups of electroencephalogram data according to channels to generate new artificial data.
As shown in fig. 6, the left electroencephalogram signal is a curve after the right motor imagery electroencephalogram signal in the randomly selected Physionet data set is normalized. It can be seen that the right hand motor imagery may be accompanied by side beta event related desynchronization and ipsilateral beta time related synchronization phenomena, especially apparent in a), b), w). After the data enhancement method is carried out on the randomly selected right-hand motor imagery electroencephalogram signals, a curve shown on the right side of fig. 6 is obtained. It can be seen that the artificial brain electrical signal with data enhancement is also accompanied by the phenomenon of synchronization of opposite side beta event correlation and synchronization of same side beta event correlation.
It follows that the data enhancement method can preserve the inherent characteristics of the electroencephalogram signal, and the effectiveness of the method can be demonstrated.
The data enhancement method provided by the invention can generate a large number of electroencephalogram artificial samples with reserved characteristics, and can perform four-classification test on 64-channel data of all subjects (109 subjects). Fig. 2 shows the average accuracy of the global model under different parameters, the abscissa indicates the number of layers of the same kind of signal superposition, the color bar indicates the multiple of data enhancement, and the blue dotted line indicates the average accuracy before data enhancement (i.e. the average accuracy without data enhancement). As can be seen from fig. 2, the data enhancement method can improve signal classification performance in the global model. When the data enhancement times are smaller than 20 and the number of the overlapping layers of the similar labels is smaller than 10, the precision is obviously improved. And the highest average accuracy of 66.73% is achieved through 20 times of data enhancement and 5 times of label superposition.
Similar to fig. 2, fig. 3 is the average accuracy after data enhancement in subject mode. As can be seen from fig. 2, the classification performance after data enhancement is significantly improved regardless of the data enhancement multiple and the number of signal superposition layers. However, as the data enhancement factor increases, the memory footprint of the code increases and the code run time increases. Thus, 10-fold data enhancement and 5-layer signal superposition are a good choice when considering computer memory and runtime. But overall, 20 times of data enhancement and 5 layers of signal superposition achieve the highest average accuracy: 76.78%.
The accuracy of the global model and the subject model trained before and after data enhancement is shown in fig. 3. As can be seen from fig. 3, the data enhancement can effectively improve the classification accuracy of the convolutional neural network algorithm. The four-class average accuracy of the global model and the subject model increased from 64.72% to 66.73% and 73.46% to 76.78%, respectively, which demonstrates the feasibility and effectiveness of the data enhancement method, especially in the subject model after introduction of the transfer learning.
In summary, the above embodiments are only preferred embodiments of the present invention, and are not intended to limit the scope of the present invention. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
Claims (3)
1. The motor imagery electroencephalogram signal classification method is characterized by comprising the following steps of:
acquiring electroencephalogram signal data of at least two types of motor imagery tasks in a Physionet data set;
performing data enhancement on the electroencephalogram data, randomly selecting more than two groups of electroencephalogram data with the same type of labels, superposing and standardizing the electroencephalogram data according to channels, namely normalizing and superposing the electroencephalogram data with the motor imagery tasks with the same type of labels to generate the electroencephalogram data;
inputting the electroencephalogram signal data into a convolutional neural network for training, thereby realizing classification of the electroencephalogram signal data;
the first layer of the convolutional neural network is a convolutional layer, convolutional operation is carried out on the electroencephalogram signal data along a time axis, and the size of output data is consistent with that of input data;
the second layer is a convolution layer, convolves the EEG signal along the EEG channel axis, and reduces the output size to half of the input;
the third layer is a maximum pooling layer for pooling the EEG signals along a time axis, the size of a kernel is 30 samples, and the step length is 15; flattening the layered structure to form 6300 single-dimensional neurons;
the last three layers are three full-connection layers, the first full-connection layer reduces the number of neurons from 6300 to 100, the second full-connection layer reduces the number of neurons from 100 to 32, the last full-connection layer is a softmax layer, and the number of neurons is reduced from 32 to the number of neurons in data to be classified;
the convolutional neural network training process adopts a transfer learning method;
filling 0 into the EEG signal data by adopting the same padding method so that the output size is the same as the input size after convolution operation;
the electroencephalogram signal data is subjected to data enhancement and then training, and specifically comprises the following steps:
normalizing the electroencephalogram data of each channel of each test for the original electroencephalogram data in the Physionet data set;
the first layer of the convolutional neural network adopts 100 filters, the size of a kernel is 30 samples, and the step length is 1;
the third layer of the convolutional neural network has a kernel size of 30 samples and a step size of 15.
2. A motor imagery electroencephalogram signal classification method as claimed in claim 1, wherein 100 filters are used for the convolutional neural network, the kernel size is 30 samples, and the step size is 1.
3. A motor imagery electroencephalogram classification method as claimed in claim 1 or claim 2, wherein the convolutional neural network employs a cost function of Adam optimiser, reLU activation and classification cross entropy.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010650831.9A CN111950366B (en) | 2020-07-08 | 2020-07-08 | Convolutional neural network motor imagery electroencephalogram classification method based on data enhancement |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010650831.9A CN111950366B (en) | 2020-07-08 | 2020-07-08 | Convolutional neural network motor imagery electroencephalogram classification method based on data enhancement |
Publications (2)
Publication Number | Publication Date |
---|---|
CN111950366A CN111950366A (en) | 2020-11-17 |
CN111950366B true CN111950366B (en) | 2023-10-24 |
Family
ID=73341419
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202010650831.9A Active CN111950366B (en) | 2020-07-08 | 2020-07-08 | Convolutional neural network motor imagery electroencephalogram classification method based on data enhancement |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN111950366B (en) |
Families Citing this family (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112883855A (en) * | 2021-02-04 | 2021-06-01 | 东北林业大学 | Electroencephalogram signal emotion recognition based on CNN + data enhancement algorithm Borderline-SMOTE |
CN113499524A (en) * | 2021-07-23 | 2021-10-15 | 华南理工大学 | Auxiliary rehabilitation training system using motor imagery electroencephalogram detection |
Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105550701A (en) * | 2015-12-09 | 2016-05-04 | 福州华鹰重工机械有限公司 | Real-time image extraction and recognition method and device |
JP2017202183A (en) * | 2016-05-12 | 2017-11-16 | 株式会社国際電気通信基礎技術研究所 | Electroencephalogram pattern classification device, electroencephalogram pattern classification method, electroencephalogram pattern classification program, and neuro-feedback system |
CN108960182A (en) * | 2018-07-19 | 2018-12-07 | 大连理工大学 | A kind of P300 event related potential classifying identification method based on deep learning |
CN109784242A (en) * | 2018-12-31 | 2019-05-21 | 陕西师范大学 | EEG Noise Cancellation based on one-dimensional residual error convolutional neural networks |
CN109784211A (en) * | 2018-12-26 | 2019-05-21 | 西安交通大学 | A kind of Mental imagery Method of EEG signals classification based on deep learning |
CN110110584A (en) * | 2019-03-14 | 2019-08-09 | 杭州电子科技大学 | A kind of emotional characteristics classification method based on CNN |
CN110929581A (en) * | 2019-10-25 | 2020-03-27 | 重庆邮电大学 | Electroencephalogram signal identification method based on space-time feature weighted convolutional neural network |
-
2020
- 2020-07-08 CN CN202010650831.9A patent/CN111950366B/en active Active
Patent Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105550701A (en) * | 2015-12-09 | 2016-05-04 | 福州华鹰重工机械有限公司 | Real-time image extraction and recognition method and device |
JP2017202183A (en) * | 2016-05-12 | 2017-11-16 | 株式会社国際電気通信基礎技術研究所 | Electroencephalogram pattern classification device, electroencephalogram pattern classification method, electroencephalogram pattern classification program, and neuro-feedback system |
CN108960182A (en) * | 2018-07-19 | 2018-12-07 | 大连理工大学 | A kind of P300 event related potential classifying identification method based on deep learning |
CN109784211A (en) * | 2018-12-26 | 2019-05-21 | 西安交通大学 | A kind of Mental imagery Method of EEG signals classification based on deep learning |
CN109784242A (en) * | 2018-12-31 | 2019-05-21 | 陕西师范大学 | EEG Noise Cancellation based on one-dimensional residual error convolutional neural networks |
CN110110584A (en) * | 2019-03-14 | 2019-08-09 | 杭州电子科技大学 | A kind of emotional characteristics classification method based on CNN |
CN110929581A (en) * | 2019-10-25 | 2020-03-27 | 重庆邮电大学 | Electroencephalogram signal identification method based on space-time feature weighted convolutional neural network |
Non-Patent Citations (1)
Title |
---|
基于卷积神经网络的运动想象脑电信号分类研究;张琳琳;《中国优秀硕士学位论文全文数据库信息科技辑》;20200516(第06期);全文 * |
Also Published As
Publication number | Publication date |
---|---|
CN111950366A (en) | 2020-11-17 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Ghaemi et al. | Automatic channel selection in EEG signals for classification of left or right hand movement in Brain Computer Interfaces using improved binary gravitation search algorithm | |
Zhao et al. | A multi-branch 3D convolutional neural network for EEG-based motor imagery classification | |
Feng et al. | An Optimized Channel Selection Method Based on Multifrequency CSP‐Rank for Motor Imagery‐Based BCI System | |
Zhang et al. | Spatiotemporal sparse Bayesian learning with applications to compressed sensing of multichannel physiological signals | |
Esfahani et al. | Classification of primitive shapes using brain–computer interfaces | |
CN114266276B (en) | Motor imagery electroencephalogram signal classification method based on channel attention and multi-scale time domain convolution | |
CN113180692B (en) | Electroencephalogram signal classification and identification method based on feature fusion and attention mechanism | |
CN111950366B (en) | Convolutional neural network motor imagery electroencephalogram classification method based on data enhancement | |
CN111783942A (en) | Brain cognition process simulation method based on convolution cyclic neural network | |
Erkilinc et al. | Camera control with EMG signals using principal component analysis and support vector machines | |
Gao et al. | Multi-ganglion ANN based feature learning with application to P300-BCI signal classification | |
Obeidat et al. | Introducing the edges paradigm: a P300 brain–computer interface for spelling written words | |
Gupta et al. | Detecting eye movements in EEG for controlling devices | |
Malekmohammadi et al. | An efficient hardware implementation for a motor imagery brain computer interface system | |
CN108710895A (en) | Motor imagery electroencephalogram signal classification method based on independent component analysis | |
Kirar et al. | Relevant frequency band selection using Sequential forward feature selection for motor imagery brain computer interfaces | |
CN117520891A (en) | Motor imagery electroencephalogram signal classification method and system | |
Wei et al. | Cross-subject EEG channel selection method for lower limb brain-computer interface | |
Fauzi et al. | Transfer learning of bci using cur algorithm | |
Zahra et al. | Exploration of pattern recognition methods for motor imagery EEG signal with convolutional neural network approach | |
Frolov et al. | Using multiple decomposition methods and cluster analysis to find and categorize typical patterns of EEG activity in motor imagery brain–computer interface experiments | |
CN111772629A (en) | Brain cognitive skill transplantation method | |
Won et al. | Selective subject pooling strategy to achieve subject-independent motor imagery BCI | |
Feng et al. | Intention recognition of upper limb movement on electroencephalogram signal based on CSP-CNN | |
Zhang et al. | Recognition method for multi-class motor imagery EEG based on channel frequency selection |
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 | ||
GR01 | Patent grant | ||
GR01 | Patent grant |