CN109598219A - A kind of adaptive electrode method for registering for robust myoelectric control - Google Patents
A kind of adaptive electrode method for registering for robust myoelectric control Download PDFInfo
- Publication number
- CN109598219A CN109598219A CN201811414282.4A CN201811414282A CN109598219A CN 109598219 A CN109598219 A CN 109598219A CN 201811414282 A CN201811414282 A CN 201811414282A CN 109598219 A CN109598219 A CN 109598219A
- Authority
- CN
- China
- Prior art keywords
- electrode
- matrix
- electromyographic
- classifier
- data
- 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.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 48
- 230000003044 adaptive effect Effects 0.000 title claims abstract description 16
- 230000003183 myoelectrical effect Effects 0.000 title claims abstract description 9
- 238000012360 testing method Methods 0.000 claims abstract description 54
- 238000012549 training Methods 0.000 claims abstract description 43
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 11
- 230000008569 process Effects 0.000 claims abstract description 7
- 239000011159 matrix material Substances 0.000 claims description 53
- 238000002567 electromyography Methods 0.000 claims description 23
- 230000000284 resting effect Effects 0.000 claims description 11
- 230000009471 action Effects 0.000 claims description 7
- 238000005516 engineering process Methods 0.000 claims description 7
- 238000004458 analytical method Methods 0.000 claims description 6
- 238000004070 electrodeposition Methods 0.000 claims description 6
- 230000000875 corresponding effect Effects 0.000 claims description 5
- 230000004118 muscle contraction Effects 0.000 claims description 3
- 238000011176 pooling Methods 0.000 claims description 3
- 238000012935 Averaging Methods 0.000 claims description 2
- 238000000605 extraction Methods 0.000 claims description 2
- 238000005070 sampling Methods 0.000 claims description 2
- 238000001228 spectrum Methods 0.000 claims description 2
- 230000036962 time dependent Effects 0.000 claims description 2
- 238000010276 construction Methods 0.000 claims 2
- 230000015572 biosynthetic process Effects 0.000 claims 1
- 238000003786 synthesis reaction Methods 0.000 claims 1
- 230000033001 locomotion Effects 0.000 abstract description 11
- 210000003811 finger Anatomy 0.000 description 5
- 230000006870 function Effects 0.000 description 5
- 230000000694 effects Effects 0.000 description 4
- 238000002474 experimental method Methods 0.000 description 4
- 238000003909 pattern recognition Methods 0.000 description 3
- 230000009286 beneficial effect Effects 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 210000003205 muscle Anatomy 0.000 description 2
- 239000013598 vector Substances 0.000 description 2
- 238000003491 array Methods 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 230000000052 comparative effect Effects 0.000 description 1
- 230000008602 contraction Effects 0.000 description 1
- 238000007796 conventional method Methods 0.000 description 1
- 210000003414 extremity Anatomy 0.000 description 1
- 210000000245 forearm Anatomy 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 210000004932 little finger Anatomy 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000004445 quantitative analysis Methods 0.000 description 1
- 210000002027 skeletal muscle Anatomy 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000013526 transfer learning Methods 0.000 description 1
- 210000000707 wrist Anatomy 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
- G06F2218/08—Feature extraction
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- 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
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/12—Classification; Matching
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Data Mining & Analysis (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Biomedical Technology (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (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 kind of adaptive electrode method for registering for robust myoelectric control, based on convolutional neural networks structure, the matching for completing regional area between training data (before electrode offset) and test data (after electrode offset), to realize the self-aligned or self calibration of electrod-array；The present invention uses complete unsupervised learning mode, is not necessarily to any additional calibration data, is done directly electrode self-calibration process in test phase.On this basis, the data for the record area that line skew poised for battle forward laps are identified, it can be ensured that the high accuracy of recognizing model of movement.
Description
Technical Field
The invention relates to the field of biological signal processing, in particular to a self-adaptive electrode registration method for robust electromyography control.
Background
The electromyographic signals are bioelectric signals accompanying muscle contraction, which contain abundant movement control information and can be used for capturing movement or movement intentions and analyzing the movement or movement intentions into control instructions in a human-computer interface. Surface electromyographic signals are collected from the skin surface and are widely used in electromyographic control because of their advantage of being non-invasive. In particular, the electromyography control technology can be applied to devices such as artificial limbs, exoskeleton robots and the like. Surface electromyogram pattern recognition is a milestone technology in the field of electromyogram control, and can realize control of multiple degrees of freedom by identifying various muscle movement patterns through a training classifier. Although this technique ideally has a high action classification rate in a laboratory, when the measurement electrode is shifted due to re-wearing in actual use, the performance of the classifier trained based on the data before shifting is greatly reduced, or even the classifier is not suitable any more, and the re-training of the classifier brings a great use load. This problem is one of the difficulties that prevent the widespread application of electromyographic pattern recognition control technology.
In previous work, attempts were made to extend all possible offset position data to the training set to enhance the robustness of the recognizer, to increase the electrode spacing to make the classifier insensitive to the electrode position, and to adaptively change the classifier with a small number of samples to fit a new feature space, but the above studies either still accompanied a severe retraining burden or had difficulty achieving the desired recognition effect. In recent years, with the widespread use of high-density array electrodes, some researchers have utilized two-dimensional spatial information in high-density array signals to address the effects of electrode offset. However, these works only use high density data to simulate electrode deflection to give quantitative analysis, and the electrode was not re-worn in the experiment, which is still significantly different from the actual application.
Disclosure of Invention
The invention aims to provide a self-adaptive electrode registration method for robust electromyography control, which greatly improves the identification accuracy of action tasks after electrode deflection when an electrode is worn again in practical application by designing a model structure based on a convolutional neural network and a unique self-calibration method.
The purpose of the invention is realized by the following technical scheme:
an adaptive electrode registration method for robust electromyography control, comprising:
constructing training data by using electromyographic signals collected by currently worn electrode equipment;
re-wearing electrode equipment, enabling the placement position of the electrode to be consistent with the previous position as much as possible, and constructing test data by utilizing the collected myoelectric signals;
a model structure based on a convolutional neural network is constructed in advance to serve as a classifier, the network is trained by utilizing training data to obtain optimal network model parameters, and a classifier model is obtained by combining the optimal network model parameters and the training data;
constructing a template by using training samples in training data, respectively matching a certain number of test samples in the test data and outputting a classification result by using a classifier, thereby initializing an electrode offset calibration matrix at the current stage;
in a later stage, the electrode positions of other test samples in the test data are registered by using the electrode offset calibration matrix, a classification result is given by a classifier, and meanwhile, the electrode offset calibration matrix is updated by using other test samples in the test data in an iterative manner.
According to the technical scheme provided by the invention, the matching of the local area between the training data (before electrode offset) and the test data (after electrode offset) is completed based on the convolutional neural network structure, so that the self-alignment or self-calibration of the electrode array is realized; the invention uses a complete unsupervised learning mode, does not need any additional calibration data, and directly finishes the self-calibration process of the electrode in the test stage. On the basis, the data of the recording areas which are overlapped before and after the array deviation are identified, and the high accuracy of the motion pattern identification can be ensured.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings needed to be used in the description of the embodiments are briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings based on the drawings without creative efforts.
Fig. 1 is a flowchart of an adaptive electrode registration method for robust electromyography control according to an embodiment of the present invention;
fig. 2 is a schematic diagram of a myoelectricity collecting device and an electrode placement position according to an embodiment of the present invention;
fig. 3 is a diagram of a neural network structure according to an embodiment of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention are clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments of the present invention without making any creative effort, shall fall within the protection scope of the present invention.
The embodiment of the invention provides an adaptive electrode registration method for robust electromyography control, which comprises the following steps of:
1. and constructing training data by using the electromyographic signals collected by the currently worn electrode equipment.
1) The electrode equipment selects a flexible high-density electrode array with p row channels, q column channels and D density. As an example, it may be provided that: p is 10, q is 10 and D is 7 mm.
2) The electrode equipment is placed at a position to be measured, continuous electromyographic signals when K action tasks are executed are collected one by one, and then the electromyographic signals in a section of resting state are collected.
For example, the forearm extensor muscle group may be selected as the acquisition object, and at this time, K is set to 6, that is, 6 gesture motions are acquired: the index finger is stretched, the middle finger is stretched, the little finger is stretched, the index finger and the middle finger are stretched together, and the last three fingers are stretched together and stretched into the wrist. Each action was taken 5 times, each time with moderate force isometric contraction held for 5 seconds, and the subjects were given sufficient rest time between exercise sessions. Myoelectric signals in a resting state of 5 seconds can be collected.
3) And segmenting the continuous electromyographic signals by using the electromyographic signals in the rest state to obtain a series of sample data.
Dividing the continuous electromyographic signal stream into a certain number of analysis windows by using a sliding window technology, wherein the window length is W (for example, W is 512ms), and the sliding increment is L (for example, L is 128 ms); selecting an analysis window formed by electromyographic signal flows in a resting state, and calculating a resting state threshold value as Th; and judging whether all analysis windows are active segments of muscle contraction or not by utilizing the resting state threshold Th, and if so, marking a corresponding action task label and taking the label as sample data.
For example, the resting state threshold Th may be obtained by adding three times of standard deviation to the average value of all the channel electromyogram signals in the sample in the resting state.
4) And performing multi-feature extraction on the electromyographic signals of each channel in each sample data, thereby constructing each sample into a preliminary two-dimensional electromyographic feature image.
The extracted multiple features comprise time domain features including waveform length WL, and f in the time-dependent power spectrum description features1And f6Characteristic; and then, constructing each sample into a preliminary two-dimensional electromyographic feature image, wherein the length is p, the width is q, and the feature number and the color channel dimension are the same.
5) And (3) independently extracting the median subregion of the preliminary two-dimensional electromyographic feature image, interpolating by S times by using an interpolation method, and adding the interpolation result to the training data.
In the embodiment of the invention, the length of the median subregion, namely the middle part, is p0Width of q0And (3) corresponding areas (defined as core recording areas) are interpolated by S times in a manner similar to the step 5), and then training data are added, wherein the length of the electromyographic feature image of the interpolated core recording area is M, and the width of the electromyographic feature image of the interpolated core recording area is N.
As an example, it may be provided that: p is a radical of0＝8，q0＝8，M＝224，N＝224。
For example, the interpolation method may be bicubic interpolation.
2. And re-wearing the electrode equipment, enabling the placement position of the electrode to be consistent with that of the electrode as far as possible, and constructing test data by utilizing the collected electromyographic signals.
The method for constructing the test data is similar to the method for constructing the training data, and the difference is that after the preliminary two-dimensional electromyographic feature image is obtained by utilizing the steps 1) to 4) mentioned in the step 1), the preliminary two-dimensional electromyographic feature image is interpolated by S times by utilizing an interpolation method, so that a high-resolution electromyographic feature image is constructed, wherein the length of the high-resolution electromyographic feature image is m, and the width of the high-resolution electromyographic feature image is n.
As an example, it may be provided that: m is 280 and n is 280.
For example, the interpolation method may be bicubic interpolation.
3. A model structure based on a convolutional neural network is constructed in advance to serve as a classifier, the network is trained by utilizing training data to obtain optimal network model parameters, and a classifier model is obtained by combining the optimal network model parameters and the training data.
In the embodiment of the invention, the underlying network structure of the model structure of the convolutional neural network is migrated from the image field to classify and identify the trained network (for example, VGG16), then a plurality of convolutional layers and pooling layers are connected, batch standardization technology is added, and the dropout technology enhances the generalization capability of the model; the full connected layer of convolution is used for realizing the sliding window method of convolution, and the last layer is a softmax layer used for classification and identification.
As shown in fig. 3, an optimal network model parameter is exemplarily given, wherein smallregion represents an electromyogram of a core registration area, conv represents a convolution layer, MaxPooling2D represents a 2-dimensional maximum pooling layer, and BatchNormalization represents a batch normalization layer. The convolutional layer block contains the size of the convolutional kernel, and the number of kernels, e.g., "3 x3conv, 64", represents that the convolutional kernel of the convolutional layer has a size of 3x3 and the number of kernels is 64.
The main process of obtaining the classifier model by combining the optimal network model parameters and the training data is as follows: and (3) jointly transmitting the core recording area electromyographic feature image of the training data and the corresponding action task label to a convolutional neural network, and training the weight of the optimal network model parameter in a gradient descent mode to obtain a classifier model.
4. And constructing a template by using the training samples in the training data, respectively matching a certain number of test samples in the test data, and outputting a classification result by using a classifier, thereby initializing the electrode offset calibration matrix at the current stage.
Constructing an electromyographic characteristic image template for core recording area electromyographic characteristic images of K action tasks in the over-training data, and aiming at the front Nc(e.g., N)c50) test samples, searching the electromyographic feature image presented by each test sample in an image matching mode by taking the electromyographic feature image template as a target; the relative position of each test sample after image registration can be obtained in the process, the category identification result of each test sample is obtained through the relative position of the image registration and the classifier result, and N is synthesizedcAnd initializing the electrode offset calibration matrix A at the moment according to the relative position and the classifier result of each test sample after image registration is completed. The specific implementation mode is as follows:
the image matching method comprises the following steps:
where(1≤i≤m-M+1,1≤j≤n-N+1,1≤c≤K)
s is a WL characteristic dimension matrix of a high-resolution myoelectric characteristic image in the test data, and m and n are image length and width of the high-resolution myoelectric characteristic image; the T matrix is a target to be matched, is derived from training data and is obtained by averaging WL characteristic dimension matrixes of all core recording area electromyographic characteristic images under the c-th category of the training data, and the length and the width of the matrix are M and N; i and j represent the image matching of the sub-region of the S matrix at the (i, j) position and the T matrix, which together identify all possible offset positions;
the prediction categories of the test samples are:
wherein,represents Dmatc hCoordinates of global minima in the matrix; the R matrix outputs the classification result for the classifier, the fix (-) function has the function of rounding the input value, dsIs the down sampling rate; illustratively, d may be sets＝7.5。
The predicted position of the electrode offset is then recorded as:
initializing a D matrix when the algorithm is started, wherein the D matrix has the same dimensionality as the R matrix; then, the output result R of the classifier of each sample is used for carrying out iterative updating, and N is passedcAfter each test sample, initializing an electrode offset calibration matrix A through the D matrix, wherein the electrode offset calibration matrix A reflects the specific direction and distance of electrode offset:
wherein D (: c) represents the whole two-dimensional matrix under the condition that the third dimension of the D matrix is the value of c; the Top5 (-) function leaves the maximum 5 values of the input matrix unchanged and zeros other values, and the max (-) function returns the maximum value of the input matrix, with ε being a set coefficient to prevent the denominator being 0, and exemplary, ε may be set to 0.001.
5. In a later stage, the electrode positions of other test samples in the test data are registered by using the electrode offset calibration matrix, a classification result is given by a classifier, and meanwhile, the electrode offset calibration matrix is updated by using other test samples in the test data in an iterative manner.
In the embodiment of the invention, the electrode positions of other test samples in the test data are registered by using the electrode offset calibration matrix, and a classification result is given by a classifier; the classification result for a certain test sample is expressed as:
wherein R (: c) represents the whole two-dimensional matrix of the R matrix under the condition that the third dimension is c value, and the sum (-) function returns the sum of all values of the input matrix.
Then, every time NuAfter 50 test samples have been input, the electrode offset calibration matrix a is continuously updated according to the method in step 4.
To illustrate the performance of the above-described protocol of the present invention, comparative experiments were conducted using conventional methods with the above-described protocol of the present invention.
In the comparison experiment, an LDA classifier is used for comparing the recognition rate change degree before and after the deviation. LDA is used as a classic classification method in the myoelectricity control field, and the classifier has good classification effect and good robustness. The method comprises the following specific steps: firstly, selecting a subject to acquire electromyographic signals through the step 1, then extracting multiple features from each channel of a sliding window sample, and then integrating each sample into a vector with the length of 300 to form training data; the training data is then used to train the LDA classifier. When the electrode is worn again and electrode offset occurs, the newly acquired electromyographic signals are converted into vectors with the length of 300 in the same way, and classification and identification are carried out by directly utilizing the LDA classifier trained before.
And measuring the identification result by using the identification accuracy, wherein the identification accuracy is the number of correctly identified samples divided by the number of all samples. The accuracy obtained for the comparison method and the method proposed by the present invention for 4 offset experiments is shown in table 1:
TABLE 1 results of comparison of accuracy
The beneficial effects of the scheme of the invention are as follows: the invention utilizes a two-dimensional high-density electrode array to collect surface electromyographic signals, and the surface electromyographic signals are regarded as the time-space mode imaging of skeletal muscle activity. The electromyographic feature image patterns associated with the task within the common coverage area of the front and rear arrays of electrode offset are assumed to be invariant. Based on the scientific hypothesis, the invention provides a method for detecting and identifying the target of the electromyographic feature image for the first time, which is used for solving the problem of motion recognition when the electromyographic electrode array is deviated. Specifically, the invention provides a convolutional neural network structure based on transfer learning, and the matching of a local area between training data (before electrode offset) and test data (after electrode offset) is searched and completed from a preprocessed high-resolution electromyographic feature image, so that the self-alignment or self-calibration of an electrode array is realized. The invention uses a complete unsupervised learning mode, does not need any additional calibration data, and directly finishes the self-calibration process of the electrode in the test stage. On the basis, the data of the recording areas which are overlapped before and after the array deviation are identified, and the high accuracy of the motion pattern identification can be ensured. The invention provides a beneficial solution for the problem of electrode array deviation in electromyographic pattern recognition.
Through the above description of the embodiments, it is clear to those skilled in the art that the above embodiments can be implemented by software, and can also be implemented by software plus a necessary general hardware platform. With this understanding, the technical solutions of the embodiments can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a usb disk, a removable hard disk, etc.), and includes several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods according to the embodiments of the present invention.
The above description is only for the preferred embodiment of the present invention, but the scope of the present invention is not limited thereto, and any changes or substitutions that can be easily conceived by those skilled in the art within the technical scope of the present invention are included in the scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.
Claims (10)
1. An adaptive electrode registration method for robust electromyography control, comprising:
constructing training data by using electromyographic signals collected by currently worn electrode equipment;
re-wearing electrode equipment, enabling the placement position of the electrode to be consistent with the previous position as much as possible, and constructing test data by utilizing the collected myoelectric signals;
a model structure based on a convolutional neural network is constructed in advance to serve as a classifier, the network is trained by utilizing training data to obtain optimal network model parameters, and a classifier model is obtained by combining the optimal network model parameters and the training data;
constructing a template by using training samples in training data, respectively matching a certain number of test samples in the test data and outputting a classification result by using a classifier, thereby initializing an electrode offset calibration matrix at the current stage;
in a later stage, the electrode positions of other test samples in the test data are registered by using the electrode offset calibration matrix, a classification result is given by a classifier, and meanwhile, the electrode offset calibration matrix is updated by using other test samples in the test data in an iterative manner.
2. The adaptive electrode registration method for robust electromyography control according to claim 1, wherein the construction of training data using electromyography signals collected by currently worn electrode devices comprises:
the electrode equipment selects a flexible high-density electrode array with p row channels, q column channels and D density;
placing electrode equipment at a position to be measured, acquiring continuous electromyographic signals one by one when executing K action tasks, and then acquiring electromyographic signals in a section of rest state;
segmenting the continuous electromyographic signals by using the electromyographic signals in the resting state to obtain a series of sample data;
performing multi-feature extraction on the electromyographic signals of each channel in each sample data, and constructing each sample into a primary two-dimensional electromyographic feature image;
and (3) independently extracting the median subregion of the preliminary two-dimensional electromyographic feature image, interpolating by S times by using an interpolation method, and adding the interpolation result to the training data.
3. The adaptive electrode registration method for robust electromyography control according to claim 1, wherein the segmenting the continuous electromyography signal with the electromyography signal at rest state to obtain a series of sample data comprises:
dividing the continuous electromyographic signal flow into a certain number of analysis windows by using a sliding window technology, wherein the window length is W, and the sliding increment is L; selecting an analysis window formed by electromyographic signal flows in a resting state, and calculating a resting state threshold value as Th; and judging whether all analysis windows are active segments of muscle contraction or not by utilizing the resting state threshold Th, and if so, marking a corresponding action task label and taking the label as sample data.
4. The adaptive electrode registration method for robust electromyography control according to claim 1, wherein the extracted multi-features include time-domain features, including waveform length WL, and time-dependent power spectrum descriptive features.
5. The adaptive electrode registration method for robust electromyography control according to claim 2, wherein during the test data construction process, a preliminary two-dimensional electromyography feature image is constructed in the same manner as when training data is constructed, and then the preliminary two-dimensional electromyography feature image is interpolated by S-fold using an interpolation method to construct a high-resolution electromyography feature image.
6. The adaptive electrode registration method for robust electromyography control according to claim 1, wherein an underlying network structure of a model structure of a convolutional neural network is migrated from an image domain to classify and recognize a trained network, followed by a plurality of convolutional layers and pooling layers, and a batch standardization technique is added; the full connected layer of convolution is used for realizing the sliding window method of convolution, and the last layer is a softmax layer used for classification and identification.
7. The adaptive electrode registration method for robust electromyography control according to claim 2, wherein deriving a classifier model combining optimal network model parameters with training data comprises:
and (3) jointly transmitting the core recording area electromyographic feature images and the corresponding action task labels in the training data to a convolutional neural network, and training the weight of the optimal network model parameter in a gradient descent mode to obtain a classifier model.
8. The adaptive electrode registration method for robust electromyography control according to claim 5, wherein the constructing a template by using the training samples in the training data, respectively matching a certain number of test samples in the test data and outputting a classification result by using a classifier, so as to initialize the electrode offset calibration matrix of the current stage comprises:
constructing an electromyographic feature image template through core recording area electromyographic feature images of K action tasks in training data, and aiming at the front NcThe method comprises the following steps that (1) each test sample is searched in an electromyographic feature image presented by each test sample in an image matching mode by taking an electromyographic feature image template as a target; in the process, for each test sample, the relative position after image registration can be obtained, and the category identification result of each test sample is obtained through the relative position of image registration and the result of a classifier;
synthesis of NcAnd initializing the electrode offset calibration matrix A at the moment according to the relative position and the classifier result of each test sample after image registration is completed.
9. The adaptive electrode registration method for robust electromyography control according to claim 8,
the image matching method comprises the following steps:
where(1≤i≤m-M+1,1≤j≤n-N+1,1≤c≤K)
s is a WL characteristic dimension matrix of a high-resolution myoelectric characteristic image in the test data, and m and n are image length and width of the high-resolution myoelectric characteristic image; the T matrix is a target to be matched, is derived from training data and is obtained by averaging WL characteristic dimension matrixes of all core recording area electromyographic characteristic images under the c-th category of the training data, and the length and the width of the matrix are M and N;
the class prediction of the test sample is:
wherein,represents Dmatc hCoordinates of global minima in the matrix; the R matrix is the output result of the classifier, the function of the fix (-) is to round the input value, dsIs the down sampling rate;
the predicted position of the electrode offset is then recorded as:
initializing a D matrix when the algorithm is started, wherein the D matrix has the same dimensionality as the R matrix; then, the output result R of the classifier of each sample is used for carrying out iterative updating, and N is passedcAfter each test sample, initializing an electrode offset calibration matrix A through the D matrix, wherein the electrode offset calibration matrix A reflects the specific direction and distance of electrode offset:
wherein D (: c) represents the whole two-dimensional matrix under the condition that the third dimension of the D matrix is the value of c; the Top5 (-) function leaves the maximum 5 values of the input matrix unchanged and zeros others, and the max (-) function returns the maximum value of the input matrix, ε is a set coefficient to prevent the denominator being 0.
10. The adaptive electrode registration method for robust electromyography control according to claim 9, wherein electrode positions of other test samples in the test data are registered by using an electrode offset calibration matrix, and a classification result is given by a classifier; the classification result for a certain test sample is expressed as:
wherein R (: c) represents the whole two-dimensional matrix of the R matrix under the condition that the third dimension is c value, and the sum (-) function returns the sum of all values of the input matrix.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811414282.4A CN109598219B (en) | 2018-11-20 | 2018-11-20 | Adaptive electrode registration method for robust electromyography control |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811414282.4A CN109598219B (en) | 2018-11-20 | 2018-11-20 | Adaptive electrode registration method for robust electromyography control |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109598219A true CN109598219A (en) | 2019-04-09 |
| CN109598219B CN109598219B (en) | 2022-09-02 |
Family
ID=65959562
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811414282.4A Active CN109598219B (en) | 2018-11-20 | 2018-11-20 | Adaptive electrode registration method for robust electromyography control |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109598219B (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110141232A (en) * | 2019-06-11 | 2019-08-20 | 中国科学技术大学 | Data enhancement methods for the identification of robust electromyography signal |
| CN111121607A (en) * | 2019-12-13 | 2020-05-08 | 深圳大学 | Method for training three-dimensional positioning model and three-dimensional positioning method and device |
| CN111401166A (en) * | 2020-03-06 | 2020-07-10 | 中国科学技术大学 | Robust gesture recognition method based on electromyographic information decoding |
| CN111657941A (en) * | 2020-07-10 | 2020-09-15 | 中国科学技术大学 | Electrode correction and myoelectric pattern recognition method based on muscle core activation region |
| CN111857349A (en) * | 2020-07-28 | 2020-10-30 | 中国科学技术大学 | Wrist strap type gesture recognition equipment with self-repairing and self-calibrating functions and method |
| CN114548165A (en) * | 2022-02-18 | 2022-05-27 | 中国科学技术大学 | Electromyographic mode classification method capable of crossing users |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101057795A (en) * | 2007-05-18 | 2007-10-24 | 天津大学 | Artificial hand using muscle electrical and electroencephalogram cooperative control and controlling method thereof |
| CN105012057A (en) * | 2015-07-30 | 2015-11-04 | 沈阳工业大学 | Intelligent artificial limb based on double-arm electromyogram and attitude information acquisition and motion classifying method |
| WO2017073770A1 (en) * | 2015-10-29 | 2017-05-04 | 国立大学法人電気通信大学 | Method of identifying myoelectric pattern employing time series information, and myoelectric prosthetic hand |
| CN106725466A (en) * | 2016-12-31 | 2017-05-31 | 安徽工业大学 | A kind of bow-backed alarm set and based reminding method based on surface myoelectric technology |
| CN108491077A (en) * | 2018-03-19 | 2018-09-04 | 浙江大学 | A kind of surface electromyogram signal gesture identification method for convolutional neural networks of being divided and ruled based on multithread |
-
2018
- 2018-11-20 CN CN201811414282.4A patent/CN109598219B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101057795A (en) * | 2007-05-18 | 2007-10-24 | 天津大学 | Artificial hand using muscle electrical and electroencephalogram cooperative control and controlling method thereof |
| CN105012057A (en) * | 2015-07-30 | 2015-11-04 | 沈阳工业大学 | Intelligent artificial limb based on double-arm electromyogram and attitude information acquisition and motion classifying method |
| WO2017073770A1 (en) * | 2015-10-29 | 2017-05-04 | 国立大学法人電気通信大学 | Method of identifying myoelectric pattern employing time series information, and myoelectric prosthetic hand |
| CN106725466A (en) * | 2016-12-31 | 2017-05-31 | 安徽工业大学 | A kind of bow-backed alarm set and based reminding method based on surface myoelectric technology |
| CN108491077A (en) * | 2018-03-19 | 2018-09-04 | 浙江大学 | A kind of surface electromyogram signal gesture identification method for convolutional neural networks of being divided and ruled based on multithread |
Non-Patent Citations (2)
| Title |
|---|
| MUHAMMAD ZIA UR REHMAN ET AL.: "Multiday EMG-Based Classification of Hand Motions with Deep Learning Techniques", 《SENSORS》 * |
| 张驰: "基于脑电信号的情绪识别方法研究", 《万方》 * |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110141232A (en) * | 2019-06-11 | 2019-08-20 | 中国科学技术大学 | Data enhancement methods for the identification of robust electromyography signal |
| CN110141232B (en) * | 2019-06-11 | 2020-10-27 | 中国科学技术大学 | Data enhancement method for robust electromyographic signal identification |
| CN111121607A (en) * | 2019-12-13 | 2020-05-08 | 深圳大学 | Method for training three-dimensional positioning model and three-dimensional positioning method and device |
| CN111121607B (en) * | 2019-12-13 | 2021-12-07 | 深圳大学 | Method for training three-dimensional positioning model and three-dimensional positioning method and device |
| CN111401166A (en) * | 2020-03-06 | 2020-07-10 | 中国科学技术大学 | Robust gesture recognition method based on electromyographic information decoding |
| CN111657941A (en) * | 2020-07-10 | 2020-09-15 | 中国科学技术大学 | Electrode correction and myoelectric pattern recognition method based on muscle core activation region |
| CN111657941B (en) * | 2020-07-10 | 2022-04-19 | 中国科学技术大学 | Electrode correction and myoelectric pattern recognition method based on muscle core activation region |
| CN111857349A (en) * | 2020-07-28 | 2020-10-30 | 中国科学技术大学 | Wrist strap type gesture recognition equipment with self-repairing and self-calibrating functions and method |
| CN111857349B (en) * | 2020-07-28 | 2022-03-01 | 中国科学技术大学 | Wrist strap type gesture recognition equipment with self-repairing and self-calibrating functions and method |
| CN114548165A (en) * | 2022-02-18 | 2022-05-27 | 中国科学技术大学 | Electromyographic mode classification method capable of crossing users |
| CN114548165B (en) * | 2022-02-18 | 2024-04-02 | 中国科学技术大学 | Myoelectricity mode classification method capable of crossing users |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109598219B (en) | 2022-09-02 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109598219B (en) | Adaptive electrode registration method for robust electromyography control | |
| Zhao et al. | Learning joint space–time–frequency features for EEG decoding on small labeled data | |
| Rabin et al. | Classification of human hand movements based on EMG signals using nonlinear dimensionality reduction and data fusion techniques | |
| Reasat et al. | Detection of inferior myocardial infarction using shallow convolutional neural networks | |
| US11179066B2 (en) | Real-time spike detection and identification | |
| KR20190128978A (en) | Method for estimating human emotions using deep psychological affect network and system therefor | |
| CN111657941B (en) | Electrode correction and myoelectric pattern recognition method based on muscle core activation region | |
| CN110991268B (en) | Depth image-based Parkinson hand motion quantization analysis method and system | |
| CN110333783B (en) | Irrelevant gesture processing method and system for robust electromyography control | |
| CN110141232B (en) | Data enhancement method for robust electromyographic signal identification | |
| Guo et al. | MCDCD: Multi-source unsupervised domain adaptation for abnormal human gait detection | |
| Zhang et al. | Human identification driven by deep CNN and transfer learning based on multiview feature representations of ECG | |
| Hoshino et al. | Comparing subject-to-subject transfer learning methods in surface electromyogram-based motion recognition with shallow and deep classifiers | |
| Fang et al. | Improve inter-day hand gesture recognition via convolutional neural network-based feature fusion | |
| Zhang et al. | Ready for use: subject-independent movement intention recognition via a convolutional attention model | |
| CN111401166A (en) | Robust gesture recognition method based on electromyographic information decoding | |
| Song et al. | Adaptive common spatial pattern for single-trial EEG classification in multisubject BCI | |
| Gao et al. | Learning invariant patterns based on a convolutional neural network and big electroencephalography data for subject-independent P300 brain-computer interfaces | |
| CN114548165B (en) | Myoelectricity mode classification method capable of crossing users | |
| Kanoga et al. | Subject transfer framework based on source selection and semi-supervised style transfer mapping for sEMG pattern recognition | |
| Li et al. | MDTL: a novel and model-agnostic transfer learning strategy for cross-subject motor imagery BCI | |
| Swain et al. | Yoga pose monitoring system using deep learning | |
| Li et al. | Cross-user gesture recognition from sEMG signals using an optimal transport assisted student-teacher framework | |
| Wang et al. | Explainable deep learning for sEMG-based similar gesture recognition: A Shapley-value-based solution | |
| CN117312985A (en) | Surface electromyographic signal similar gesture recognition method based on interpretable deep learning |
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 |