CN114343679A - Surface electromyogram signal upper limb action recognition method and system based on transfer learning - Google Patents
Surface electromyogram signal upper limb action recognition method and system based on transfer learning Download PDFInfo
- Publication number
- CN114343679A CN114343679A CN202111596704.6A CN202111596704A CN114343679A CN 114343679 A CN114343679 A CN 114343679A CN 202111596704 A CN202111596704 A CN 202111596704A CN 114343679 A CN114343679 A CN 114343679A
- Authority
- CN
- China
- Prior art keywords
- data set
- upper limb
- transfer learning
- alexnet network
- electromyographic signals
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 52
- 230000009471 action Effects 0.000 title claims abstract description 43
- 210000001364 upper extremity Anatomy 0.000 title claims abstract description 39
- 238000013526 transfer learning Methods 0.000 title claims abstract description 31
- 238000012549 training Methods 0.000 claims abstract description 35
- 238000013145 classification model Methods 0.000 claims abstract description 24
- 238000012360 testing method Methods 0.000 claims abstract description 23
- 239000011159 matrix material Substances 0.000 claims abstract description 19
- 238000007781 pre-processing Methods 0.000 claims abstract description 14
- 238000003384 imaging method Methods 0.000 claims abstract description 11
- 230000033001 locomotion Effects 0.000 claims description 26
- 210000000707 wrist Anatomy 0.000 claims description 17
- 238000000605 extraction Methods 0.000 claims description 10
- 238000002567 electromyography Methods 0.000 claims description 8
- 230000001052 transient effect Effects 0.000 claims description 6
- 238000004422 calculation algorithm Methods 0.000 claims description 4
- 238000006243 chemical reaction Methods 0.000 claims description 4
- 230000005012 migration Effects 0.000 claims description 4
- 238000013508 migration Methods 0.000 claims description 4
- 238000010606 normalization Methods 0.000 claims description 4
- 230000008859 change Effects 0.000 claims description 3
- 238000002372 labelling Methods 0.000 claims description 3
- 238000013507 mapping Methods 0.000 claims description 3
- 210000002569 neuron Anatomy 0.000 claims description 3
- 230000000875 corresponding effect Effects 0.000 description 19
- 230000000694 effects Effects 0.000 description 7
- 230000003183 myoelectrical effect Effects 0.000 description 7
- 238000004458 analytical method Methods 0.000 description 5
- 238000013135 deep learning Methods 0.000 description 5
- 238000013527 convolutional neural network Methods 0.000 description 4
- 238000002474 experimental method Methods 0.000 description 4
- 239000000284 extract Substances 0.000 description 4
- 230000009286 beneficial effect Effects 0.000 description 3
- 210000003414 extremity Anatomy 0.000 description 3
- 230000007547 defect Effects 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 238000003062 neural network model Methods 0.000 description 2
- 238000007637 random forest analysis Methods 0.000 description 2
- 230000009467 reduction Effects 0.000 description 2
- 238000012706 support-vector machine Methods 0.000 description 2
- 238000009966 trimming Methods 0.000 description 2
- 238000007792 addition Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 239000003814 drug Substances 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000008014 freezing Effects 0.000 description 1
- 238000007710 freezing Methods 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 210000003205 muscle Anatomy 0.000 description 1
- 210000001087 myotubule Anatomy 0.000 description 1
- 230000002232 neuromuscular Effects 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 210000000623 ulna Anatomy 0.000 description 1
Images
Landscapes
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The invention discloses a surface electromyogram signal upper limb action recognition method and system based on transfer learning, wherein the related surface electromyogram signal upper limb action recognition method based on transfer learning comprises the following steps: s1, collecting surface electromyographic signals corresponding to upper limb actions, and preprocessing the collected surface electromyographic signals; s2, extracting features of the preprocessed surface electromyographic signals by a sliding windowing method, constructing a numerical matrix by using the extracted features, generating an image by two-dimensional imaging, expanding an image data set by turning over, and dividing the expanded image data set into a training data set and a test data set; s3, calling an Alexnet network through transfer learning, finely adjusting the Alexnet network, inputting a training data set into the finely adjusted Alexnet network for training to obtain a classification model; and S4, inputting the test data set into a classification model, and outputting a classification result by the classification model.
Description
Technical Field
The invention relates to the technical field of myoelectric artificial limbs, in particular to a surface myoelectric signal upper limb action recognition method and system based on transfer learning.
Background
Surface electromyography (surface electromyography) is a bioelectrical signal closely related to neuromuscular activity, which is the superposition of Motor unit potentials (Motor unit activity potentials) in multiple muscle fibers, both temporally and spatially. Different limb movements involve different muscle coordination patterns, and the generated surface electromyographic signals are different. The corresponding action can be distinguished by decoding the surface electromyographic signals, and the surface electromyographic signals are widely applied to the fields of clinical medicine, biomedicine, sports rehabilitation and the like in addition to the mature noninvasive acquisition technology.
In most of the prior art, the features extracted from the surface electromyogram signals are often directly used as identification parameters to identify the limb actions. And extracting characteristic information on a time domain/a frequency domain from the collected electromyographic signals in a sliding windowing manner, then assigning corresponding action labels, and putting the action labels into a pre-trained model for identification and classification. Due to the large difference among users, the model trained by using part of experimental object data has poor effect on the test of new users, namely the robustness of the model is low. In actual scene application, making a wrong judgment or instruction can generate a great risk.
With the rise of deep learning, various deep neural network models are generated at the same time. The method utilizes transfer learning, calls the existing neural network model and can be directly applied to electromyographic signal action recognition after fine adjustment. The method can simplify the calculated amount, improve the operation efficiency and effectively solve the problem of poor robustness caused by user difference.
Disclosure of Invention
Aiming at the defects of the prior art, the invention provides a surface electromyogram signal upper limb action recognition method and system based on transfer learning, and the problem of model robustness reduction caused by user difference can be effectively solved by utilizing the strong learning and analysis capability of computer vision. The method combines the advantages of a time domain feature extraction method and deep learning, converts an input form from a one-dimensional surface electromyogram time sequence signal into a two-dimensional image, extracts depth feature information by calling a multilayer convolutional neural network model, and finally obtains a classification result. The experimental result shows that the method has higher classification precision when being used for cross-human upper limb action recognition, and has better recognition effect compared with the traditional classification method. The method can effectively reduce the influence of the user difference on the model robustness, and is beneficial to improving the universality of the myoelectric artificial hand and the reality of the application of a real scene.
In order to achieve the purpose, the invention adopts the following technical scheme:
a surface electromyogram signal upper limb action recognition method based on transfer learning comprises the following steps:
s1, collecting surface electromyographic signals corresponding to upper limb actions, and preprocessing the collected surface electromyographic signals;
s2, extracting features of the preprocessed surface electromyographic signals by a sliding windowing method, constructing a numerical matrix by using the extracted features, generating an image by two-dimensional imaging, expanding an image data set by turning over, and dividing the expanded image data set into a training data set and a test data set;
s3, calling an Alexnet network through transfer learning, finely adjusting the Alexnet network, inputting a training data set into the finely adjusted Alexnet network for training to obtain a classification model;
and S4, inputting the test data set into a classification model, and outputting a classification result by the classification model.
Further, the step S1 specifically includes:
s11, synchronously acquiring surface electromyographic signals corresponding to upper limb actions according to given action instructions, and labeling corresponding to the actions;
s12, preprocessing the collected surface electromyographic signals; wherein the preprocessing comprises removing unwanted noise in the surface electromyography signal.
Further, the collecting manner in step S11 is as follows: each gesture corresponding to the upper limb movement lasts for 10 seconds, so that the surface electromyographic signals simultaneously contain transient and steady-state signals within 10 seconds, and the transient signals are eliminated by removing the electric signals of the first 5 seconds to obtain the steady-state signals.
Further, the step S2 of extracting the characteristics of the surface myoelectric signal specifically includes: setting the fixed window to be 256ms and the sliding window to be 64ms, and extracting the average absolute value, the waveform length, the number of zero-crossing points and the number of slope polarity changes of the time domain characteristics.
Further, the constructing of the numerical matrix in step S2 specifically includes: setting a fixed window to be 50ms, setting a sliding window to be 10ms, combining the extracted average absolute value, waveform length, zero-crossing point number and slope polarity change number to obtain a numerical matrix of 100x32, wherein each matrix is characterized by 50x 16;
further, the two-dimensional imaging in step S2 specifically includes: and respectively normalizing the average absolute value, the waveform length, the zero crossing point number and the slope polarity change number by a maximum and minimum normalization method to obtain a corresponding image through gray level conversion and RGB mapping.
Further, in step S3, invoking an Alexnet network through migration learning, and fine-tuning the Alexnet network specifically includes: and calling the original Alexnet network by using a transfer learning method, and modifying the number of the neurons on the fc8 layer, the softmax layer and the classoutput layer.
Further, the parameters of the Alexnet network are set as follows: the optimizer algorithm selects sgdm with an initial learning rate set to 0.002, a learning rate decay factor set to 0.5, a learning rate decay period set to 2, a maximum number of iterations set to 4, and MiniBatchSize set to 64.
Further, the motion commands given in the step S11 include hand rest, hand open, hand close, wrist flexion, wrist extension, wrist pronation, wrist supination, ulna flexion, radius flexion, fine pinch grip, key space, ball grip, and cylindrical grip.
Correspondingly, still provide surface electromyogram signal upper limbs action recognition system based on transfer learning, include:
the acquisition module is used for acquiring surface electromyographic signals corresponding to the actions of the upper limbs and preprocessing the acquired surface electromyographic signals;
the characteristic extraction module is used for extracting the characteristics of the preprocessed surface electromyographic signals by a sliding windowing method, constructing a numerical matrix by using the extracted characteristics, generating an image by two-dimensional imaging, expanding an image data set by turning over operation, and dividing the expanded image data set into a training data set and a test data set;
the training module is used for calling the Alexnet network through transfer learning, finely adjusting the Alexnet network, inputting a training data set into the finely adjusted Alexnet network for training to obtain a classification model;
and the classification module is used for inputting the test data set into the classification model, and the classification model outputs a classification result.
Compared with the prior art, the method and the device can effectively solve the problem of reduced model robustness caused by user difference by utilizing the strong learning and analyzing capability of computer vision. The method combines the advantages of a time domain feature extraction method and deep learning, converts an input form from a one-dimensional surface electromyogram time sequence signal into a two-dimensional image, extracts depth feature information by calling a multilayer convolutional neural network model, and finally obtains a classification result. The experimental result shows that the method has higher classification precision when being used for cross-human upper limb action recognition, and has better recognition effect compared with the traditional classification method. The method can effectively reduce the influence of the user difference on the model robustness, and is beneficial to improving the universality of the myoelectric artificial hand and the reality of the application of a real scene.
Drawings
Fig. 1 is a flowchart of a surface electromyogram signal upper limb movement recognition method based on transfer learning according to a first embodiment;
FIG. 2 is a schematic diagram of thirteen gestures provided in the first embodiment;
FIG. 3 is a diagram of an Alexnet network architecture after trimming as provided by one embodiment;
FIG. 4 is a schematic view of a portion of a continuous sample of a day of a subject provided in accordance with one embodiment;
FIG. 5 is a comparison chart of six wrist movements recognition experiments provided by the first embodiment;
FIG. 6 is a comparison graph of thirteen actions identified according to the first embodiment.
Detailed Description
The embodiments of the present invention are described below with reference to specific embodiments, and other advantages and effects of the present invention will be easily understood by those skilled in the art from the disclosure of the present specification. The invention is capable of other and different embodiments and of being practiced or of being carried out in various ways, and its several details are capable of modification in various respects, all without departing from the spirit and scope of the present invention. It is to be noted that the features in the following embodiments and examples may be combined with each other without conflict.
The invention aims to provide a surface electromyogram signal upper limb action recognition method and system based on transfer learning, aiming at the defects of the prior art.
Example one
The present embodiment provides a surface electromyogram signal upper limb movement recognition method based on transfer learning, as shown in fig. 1, including:
s1, collecting surface electromyographic signals corresponding to upper limb actions, and preprocessing the collected surface electromyographic signals;
s2, extracting features of the preprocessed surface electromyographic signals by a sliding windowing method, constructing a numerical matrix by using the extracted features, generating an image by two-dimensional imaging, expanding an image data set by turning over, and dividing the expanded image data set into a training data set and a test data set;
s3, calling an Alexnet network through transfer learning, finely adjusting the Alexnet network, inputting a training data set into the finely adjusted Alexnet network for training to obtain a classification model;
and S4, inputting the test data set into a classification model, and outputting a classification result by the classification model.
In step S1, a surface electromyogram signal corresponding to an upper limb movement is acquired, and the acquired surface electromyogram signal is preprocessed. The method specifically comprises the following steps:
s11, synchronously acquiring surface electromyographic signals corresponding to upper limb actions according to given action instructions, and labeling corresponding to the actions;
and in the signal acquisition stage, according to given action instructions (13 gestures in total for finger/wrist actions and 6 subjects in total), corresponding surface electromyographic signals are synchronously acquired, and labels corresponding to actions are given. The 13 collected actions are repeated twice each time, the action lasts for 5s, the rest interval between the two actions is 3s, each gesture lasts for 10 s, and the surface electromyogram signal contains transient and steady-state signals within 10 s. Transient signals were excluded by removing the first 5 seconds of signal, and only steady state signals remained in the data set.
The thirteen gestures include three basic motions, six wrist motions, and four grasping manners, as shown in fig. 2.
Three basic actions: hand Rest (HR), Hand Open (HO), and Hand Closed (HC); six wrist actions: wrist Flexion (WF), Wrist Extension (WE), Wrist Pronation (WP), Wrist Supination (WS), Ulnar Flexion (UF) and Radius Flexion (RF); four gripping ways: fine grip (FP), Key Pitch (KP), ball grip (SG) and Cylindrical Grip (CG).
In the embodiment, data acquisition is carried out on a plurality of healthy people, in each acquisition process, a subject performs hand motion according to a randomly given motion instruction, equipment synchronously acquires corresponding surface electromyographic signals, and later-stage manual work is carried out and labels corresponding to the motion are given.
S12, preprocessing the collected surface electromyographic signals; wherein the preprocessing comprises removing unwanted noise in the surface electromyography signal.
And in the signal preprocessing stage, the acquired surface electromyogram signals are preprocessed, and unnecessary noise is suppressed and eliminated through a Sallen-Key band-pass filter of 20-500Hz, a 50Hz wave trap and the like, so that the unnecessary noise of the signals is removed as much as possible.
In step S2, feature extraction is performed on the preprocessed surface electromyogram signal by a sliding windowing method, a numerical matrix is constructed using the extracted features, a two-dimensional imaging is performed to generate an image, an image data set is expanded by a flipping operation, and the expanded image data set is divided into a training data set and a test data set.
Extracting the time domain characteristics of the surface electromyographic signals: the fixed window is set to 256ms and the sliding window is set to 64ms, four types of time domain feature Mean Absolute Value (MAV), Waveform Length (WL), number of Zero Crossings (ZC), and number of slope polarity changes (SSC) are extracted, wherein in the extraction for two types of features of ZC and SSC, the threshold ε is set to 8mv to avoid the influence of low noise.
A numerical matrix construction stage: the fixed window is set to 50ms and the sliding window is set to 10ms, each matrix with characteristics of 50x16 (50 indicates 50 consecutive eigenvalues; 16 indicates the number of channels), and the four matrices are combined to obtain a numerical matrix of 100x 32.
A two-dimensional imaging stage: firstly, respectively carrying out normalization of [0 and 255] on 4 types of features by using a maximum and minimum normalization method, then converting a numerical matrix into a matrix of [0 and 1] through gray scale conversion, and finally obtaining a corresponding image by using RGB mapping, wherein the image size is set to be 227x 227.
A data set division stage: the data volume is expanded by a factor of 2 by using a flip operation. In this embodiment, the data of one subject is selected as a test set, the data of the other five subjects is selected as a training set, and each subject is sequentially used as a test set to perform a classification experiment.
In step S3, an Alexnet network is called through transfer learning, the Alexnet network is fine-tuned, and a training data set is input to the fine-tuned Alexnet network for training, so as to obtain a classification model, where the classification model is a model for gesture classification.
An original Alexnet network model is called firstly by using a model-based transfer learning method, and the model is trained by using an ImageNet data set, wherein the ImageNet data set comprises 1000 image categories. Freezing parameter information (including weight and bias) in the previous 22 layers, changing the number of neurons on the 'fc 8' fully-connected layer to n, wherein n corresponds to the number of action types, changing the corresponding subsequent 'softmax' layer and 'classoutput' layer, and keeping the parameter information of the rest layers unchanged. The framework in the original Alexnet network model and the parameter information obtained from the learning and training in the Imagenet data set are migrated to the surface myoelectric signal gesture recognition problem. Fig. 3 shows a structure of the Alexnet network after trimming.
In this embodiment, in the network parameter setting stage, the optimizer algorithm selects 'sgdm', the initial learning rate is set to 0.002, the learning rate attenuation factor is set to 0.5, the learning rate attenuation period is set to 2, the maximum number of iterations is set to 4, and MiniBatchSize is set to 64; and in the network training stage, a training set is used as input, and training is performed according to the parameter setting.
Data from one subject was selected as the test set, and data from the remaining 5 subjects was the training set. And inputting the training set into the network model after fine tuning for training.
In step S4, the test data set is input to the classification model, which outputs the classification result.
And placing the test set into the trained network for testing to obtain a classification result. The data of each subject was used as a test set to perform N experiments and then the results were averaged. And finally comparing with the traditional classification method.
Fig. 4 shows a partial continuous sample of a day for a subject. (a) And (b) pictures generated by two consecutive division of the slices by the above-described one-dimensional time-series signal 1D-two-dimensional image 2D conversion method.
FIG. 5 shows a comparison graph of six wrist motion recognition experiments (mean absolute value + standard deviation), including the results of 5 conventional classification methods, in which KNN (K-Nearest Neighbor) is the Nearest Neighbor algorithm, SVM (rbf) (support vector machine with gaussian radial basis function kernel) is the support vector machine with gaussian radial basis kernel, RF (random forest) is the random forest, and DAC (vibration Analysis Claifier) is the Discriminant Analysis method and LDA (Linear Discriminant Analysis) is the Linear Discriminant Analysis method.
Fig. 6 shows the comparison graph (mean absolute value) of thirteen motion recognitions, which includes the results of 5 conventional classification methods.
The embodiment utilizes the strong learning and analyzing capability of computer vision, and can effectively solve the problem of model robustness reduction caused by user difference. The method combines the advantages of a time domain feature extraction method and deep learning, converts an input form from a one-dimensional surface electromyogram time sequence signal into a two-dimensional image, extracts depth feature information by calling a multilayer convolutional neural network model, and finally obtains a classification result. The experimental result shows that the method has higher classification precision when being used for cross-human upper limb action recognition, and has better recognition effect compared with the traditional classification method. The method can effectively reduce the influence of the user difference on the model robustness, and is beneficial to improving the universality of the myoelectric artificial hand and the reality of the application of a real scene.
Example two
The embodiment provides a surface electromyogram signal upper limb movement recognition system based on transfer learning, which includes:
the acquisition module is used for acquiring surface electromyographic signals corresponding to the actions of the upper limbs and preprocessing the acquired surface electromyographic signals;
the characteristic extraction module is used for extracting the characteristics of the preprocessed surface electromyographic signals by a sliding windowing method, constructing a numerical matrix by using the extracted characteristics, generating an image by two-dimensional imaging, expanding an image data set by turning over operation, and dividing the expanded image data set into a training data set and a test data set;
the training module is used for calling the Alexnet network through transfer learning, finely adjusting the Alexnet network, inputting a training data set into the finely adjusted Alexnet network for training to obtain a classification model;
and the classification module is used for inputting the test data set into the classification model, and the classification model outputs a classification result.
It should be noted that the surface electromyogram signal upper limb movement recognition system based on the transfer learning provided in this embodiment is similar to the embodiment, and is not described herein again.
Compared with the prior art, the embodiment can effectively solve the problem of reduced model robustness caused by user difference by utilizing the strong learning and analyzing capability of computer vision. The method combines the advantages of a time domain feature extraction method and deep learning, converts an input form from a one-dimensional surface electromyogram time sequence signal into a two-dimensional image, extracts depth feature information by calling a multilayer convolutional neural network model, and finally obtains a classification result.
The specific embodiments described herein are merely illustrative of the spirit of the invention. Various modifications or additions may be made to the described embodiments or alternatives may be employed by those skilled in the art without departing from the spirit or ambit of the invention as defined in the appended claims.
Claims (10)
1. A surface electromyogram signal upper limb action recognition method based on transfer learning is characterized by comprising the following steps:
s1, collecting surface electromyographic signals corresponding to upper limb actions, and preprocessing the collected surface electromyographic signals;
s2, extracting features of the preprocessed surface electromyographic signals by a sliding windowing method, constructing a numerical matrix by using the extracted features, generating an image by two-dimensional imaging, expanding an image data set by turning over, and dividing the expanded image data set into a training data set and a test data set;
s3, calling an Alexnet network through transfer learning, finely adjusting the Alexnet network, inputting a training data set into the finely adjusted Alexnet network for training to obtain a classification model;
and S4, inputting the test data set into a classification model, and outputting a classification result by the classification model.
2. The method for recognizing the surface electromyogram signal upper limb movement based on the transfer learning of claim 1, wherein the step S1 specifically comprises:
s11, synchronously acquiring surface electromyographic signals corresponding to upper limb actions according to given action instructions, and labeling corresponding to the actions;
s12, preprocessing the collected surface electromyographic signals; wherein the preprocessing comprises removing unwanted noise in the surface electromyography signal.
3. The method for recognizing the surface electromyogram signal upper limb movement based on the transfer learning of claim 2, wherein the collecting manner in the step S11 is as follows: each gesture corresponding to the upper limb movement lasts for 10 seconds, so that the surface electromyographic signals simultaneously contain transient and steady-state signals within 10 seconds, and the transient signals are eliminated by removing the electric signals of the first 5 seconds to obtain the steady-state signals.
4. The method for recognizing the surface electromyography-based upper limb movement according to claim 1, wherein the step S2, the performing of the feature extraction on the surface electromyography signal specifically comprises: setting the fixed window to be 256ms and the sliding window to be 64ms, and extracting the average absolute value, the waveform length, the number of zero-crossing points and the number of slope polarity changes of the time domain characteristics.
5. The method for recognizing the surface electromyogram signal upper limb movement based on the transfer learning of claim 4, wherein the constructing of the numerical matrix in the step S2 specifically comprises: the fixed window is set to be 50ms, the sliding window is set to be 10ms, each matrix with the characteristics of 50x16 is obtained, and the extracted average absolute value, the extracted waveform length, the number of zero-crossing points and the number of slope polarity changes are combined to obtain a numerical matrix of 100x 32.
6. The method for recognizing the surface electromyogram signal upper limb movement based on the transfer learning of claim 2, wherein the two-dimensional imaging in the step S2 is specifically as follows: and respectively normalizing the average absolute value, the waveform length, the zero crossing point number and the slope polarity change number by a maximum and minimum normalization method to obtain a corresponding image through gray level conversion and RGB mapping.
7. The method for recognizing the surface electromyogram signal upper limb movement based on the migration learning according to claim 1, wherein the invoking of the Alexnet network in step S3 through the migration learning and the fine tuning of the Alexnet network are specifically: and calling the original Alexnet network by using a transfer learning method, and modifying the number of the neurons on the fc8 layer, the softmax layer and the classoutput layer.
8. The method for recognizing the surface electromyogram signal upper limb movement based on the transfer learning of claim 7, wherein the parameters of the Alexnet network are set as follows: the optimizer algorithm selects sgdm with an initial learning rate set to 0.002, a learning rate decay factor set to 0.5, a learning rate decay period set to 2, a maximum number of iterations set to 4, and MiniBatchSize set to 64.
9. The method for identifying surface electromyography-based upper limb movement of claim 2, wherein the movement commands given in step S11 include hand rest, hand open, hand close, wrist flexion, wrist extension, wrist pronation, wrist supination, ulnar flexion, radius flexion, fine pinch grip, key pitch, ball grip, and cylindrical grip.
10. Surface electromyogram signal upper limbs action identification system based on migration learning, its characterized in that includes:
the acquisition module is used for acquiring surface electromyographic signals corresponding to the actions of the upper limbs and preprocessing the acquired surface electromyographic signals;
the characteristic extraction module is used for extracting the characteristics of the preprocessed surface electromyographic signals by a sliding windowing method, constructing a numerical matrix by using the extracted characteristics, generating an image by two-dimensional imaging, expanding an image data set by turning over operation, and dividing the expanded image data set into a training data set and a test data set;
the training module is used for calling the Alexnet network through transfer learning, finely adjusting the Alexnet network, inputting a training data set into the finely adjusted Alexnet network for training to obtain a classification model;
and the classification module is used for inputting the test data set into the classification model, and the classification model outputs a classification result.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111596704.6A CN114343679A (en) | 2021-12-24 | 2021-12-24 | Surface electromyogram signal upper limb action recognition method and system based on transfer learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111596704.6A CN114343679A (en) | 2021-12-24 | 2021-12-24 | Surface electromyogram signal upper limb action recognition method and system based on transfer learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114343679A true CN114343679A (en) | 2022-04-15 |
Family
ID=81100686
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111596704.6A Pending CN114343679A (en) | 2021-12-24 | 2021-12-24 | Surface electromyogram signal upper limb action recognition method and system based on transfer learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114343679A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115169407A (en) * | 2022-07-18 | 2022-10-11 | 中国科学院国家空间科学中心 | Weak transient signal detection method and system based on time domain depth feature learning |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106980367A (en) * | 2017-02-27 | 2017-07-25 | 浙江工业大学 | A kind of gesture identification method based on myoelectricity topographic map |
| CN108388348A (en) * | 2018-03-19 | 2018-08-10 | 浙江大学 | A kind of electromyography signal gesture identification method based on deep learning and attention mechanism |
| CN112006678A (en) * | 2020-09-10 | 2020-12-01 | 齐鲁工业大学 | Electrocardiogram abnormity identification method and system based on combination of AlexNet and transfer learning |
| CN113505822A (en) * | 2021-06-30 | 2021-10-15 | 中国矿业大学 | Multi-scale information fusion upper limb action classification method based on surface electromyographic signals |
-
2021
- 2021-12-24 CN CN202111596704.6A patent/CN114343679A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106980367A (en) * | 2017-02-27 | 2017-07-25 | 浙江工业大学 | A kind of gesture identification method based on myoelectricity topographic map |
| CN108388348A (en) * | 2018-03-19 | 2018-08-10 | 浙江大学 | A kind of electromyography signal gesture identification method based on deep learning and attention mechanism |
| CN112006678A (en) * | 2020-09-10 | 2020-12-01 | 齐鲁工业大学 | Electrocardiogram abnormity identification method and system based on combination of AlexNet and transfer learning |
| CN113505822A (en) * | 2021-06-30 | 2021-10-15 | 中国矿业大学 | Multi-scale information fusion upper limb action classification method based on surface electromyographic signals |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115169407A (en) * | 2022-07-18 | 2022-10-11 | 中国科学院国家空间科学中心 | Weak transient signal detection method and system based on time domain depth feature learning |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Zhang et al. | HeartID: A multiresolution convolutional neural network for ECG-based biometric human identification in smart health applications | |
| Shen et al. | Movements classification of multi-channel sEMG based on CNN and stacking ensemble learning | |
| CN107736894A (en) | A kind of electrocardiosignal Emotion identification method based on deep learning | |
| CN110555468A (en) | Electroencephalogram signal identification method and system combining recursion graph and CNN | |
| CN111407243B (en) | Pulse signal pressure identification method based on deep learning | |
| Luque et al. | End-to-end photopleth YsmographY (PPG) based biometric authentication by using convolutional neural networks | |
| Erkilinc et al. | Camera control with EMG signals using principal component analysis and support vector machines | |
| Ye et al. | Optimal feature selection for EMG-based finger force estimation using LightGBM model | |
| Gautam et al. | Locomo-net: a low-complex deep learning framework for sEMG-based hand movement recognition for prosthetic control | |
| CN112488002B (en) | Emotion recognition method and system based on N170 | |
| CN109498370A (en) | Joint of lower extremity angle prediction technique based on myoelectricity small echo correlation dimension | |
| CN113111831A (en) | Gesture recognition technology based on multi-mode information fusion | |
| Li et al. | EEG signal classification method based on feature priority analysis and CNN | |
| Lin et al. | A normalisation approach improves the performance of inter-subject sEMG-based hand gesture recognition with a ConvNet | |
| CN114343679A (en) | Surface electromyogram signal upper limb action recognition method and system based on transfer learning | |
| CN116910464A (en) | Myoelectric signal prosthetic hand control system and method | |
| Yu et al. | The research of sEMG movement pattern classification based on multiple fused wavelet function | |
| CN110604578A (en) | Human hand and hand motion recognition method based on SEMG | |
| Bhalerao et al. | Automatic detection of motor imagery EEG signals using swarm decomposition for robust BCI systems | |
| CN115281676A (en) | Fatigue detection method based on GRU neural network and ECG signal | |
| Bo et al. | Hand gesture recognition using semg signals based on cnn | |
| CN114569143A (en) | Myoelectric gesture recognition method based on attention mechanism and multi-feature fusion | |
| Vijayvargiya et al. | Implementation of machine learning algorithms for automated human gait activity recognition using sEMG signals | |
| CN114098768A (en) | Cross-individual surface electromyographic signal gesture recognition method based on dynamic threshold and easy TL | |
| CN114626405A (en) | Real-time identity recognition method and device based on electromyographic signals and electronic equipment |
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 |