CN110389663B - sEMG gesture recognition method based on wavelet width learning system - Google Patents
sEMG gesture recognition method based on wavelet width learning system Download PDFInfo
- Publication number
- CN110389663B CN110389663B CN201910548939.4A CN201910548939A CN110389663B CN 110389663 B CN110389663 B CN 110389663B CN 201910548939 A CN201910548939 A CN 201910548939A CN 110389663 B CN110389663 B CN 110389663B
- Authority
- CN
- China
- Prior art keywords
- semg
- signal
- gesture
- learning system
- width learning
- 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
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
- G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
-
- 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
- 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
- G06F3/017—Gesture based interaction, e.g. based on a set of recognized hand gestures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/20—Movements or behaviour, e.g. gesture recognition
- G06V40/28—Recognition of hand or arm movements, e.g. recognition of deaf sign language
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- General Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Human Computer Interaction (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Evolutionary Biology (AREA)
- Psychiatry (AREA)
- Social Psychology (AREA)
- Multimedia (AREA)
- Biomedical Technology (AREA)
- Dermatology (AREA)
- Neurology (AREA)
- Neurosurgery (AREA)
- Probability & Statistics with Applications (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a sEMG gesture recognition method based on a wavelet width learning system, which comprises the following steps: step one, defining the number d of gesture action types according to corresponding recognition scenes; step two, an electromyographic signal acquisition device is used for acquiring sEMG signals s; step three, filtering and denoising the sEMG original signal according to the frequency characteristic of the sEMG signal, and removing noise outside the frequency band of 10 Hz-500 Hz by using a Butterworth filter; detecting an active segment in the sEMG signal by adopting a moving window method; step five, extracting the characteristics of the detected movable section; compared with the algorithm based on the traditional deep learning network, the method can complete training of the model and determination of parameters more quickly, so that the working efficiency is improved; nodes can be dynamically expanded to increase the recognition rate of the system without the need to fully re-build and train the model.
Description
Technical Field
The invention relates to the technical field of machine learning and signal classification, in particular to a sEMG gesture recognition method based on a wavelet width learning system.
Background
Gesture actions in body language play an important role in daily communications, such as referee actions on football stadium, etc. Therefore, scholars have focused on how to allow computers and machines to recognize gesture actions by human beings and execute corresponding programs with high efficiency and high precision. This will change the form of human-machine communication.
Currently, the recognition algorithms for gesture actions are mainly divided into the following types: a classification algorithm based on visual image recognition and a gesture recognition algorithm based on surface electromyographic signals (sEMG); the former processes images, and the method has high requirements on the image pickup equipment and high price of the equipment, so that the method is difficult to popularize, but the recognition rate has been researched into high recognition rate; while gesture recognition algorithms based on surface electromyographic signals (sEMG) have lower hardware requirements and are easy to implement than the former, the hardware for acquiring sEMG signals can be worn on the body and is not limited to a recognition product with a fixed camera position. The existing methods are not based on artificial intelligence algorithms at present to study how to classify signals.
In recent years, a great deal of research on gesture recognition algorithms is being carried out at home and abroad, including: a study of signal acquisition, a study of feature extraction and calculation, a study of signal filtering, a study of a feature classification model, and the like. Currently, the commonly used classification models are all based on classical machine learning algorithms, such as random forest networks, BP neural networks, convolutional neural networks, etc. However, these classical network models consume a significant amount of time in the training process.
Disclosure of Invention
The invention aims to overcome the defects and shortcomings of the prior art and provides a sEMG gesture recognition method based on a wavelet width learning system, which can complete training of a model and determination of parameters more quickly, so that the working efficiency is improved.
The aim of the invention is achieved by the following technical scheme:
a sEMG gesture recognition method based on a wavelet width learning system comprises the following steps:
step one, defining the number d of gesture action types according to corresponding recognition scenes, and distributing a serial number to each gesture;
step two, an electromyographic signal acquisition device is used for acquiring sEMG signals s;
step three, filtering and denoising the sEMG original signal according to the frequency characteristic of the sEMG signal, and removing noise outside the frequency band of 10 Hz-500 Hz by using a Butterworth filter at the same time:
where N is the order, ω of the filter c Filtering for a cut-off frequency;
detecting an active segment in the sEMG signal by adopting a moving window method;
step five, the detected movable section is processedExtracting row characteristics; calculating a characteristic of the signal segment, the characteristic comprising: average absolute value q 1 Root mean square q 2 Median frequency q 3 Mean frequency q 4 … … each feature of the tandem active segment is a feature vector:
x k =[q 1 ,q 2 ,...,q n ] T ,
wherein k is the kth active signal segment;
further, taking the characteristic vector of each active segment as an input vector x of the wavelet width learning classification system;
step six, designing input and output nodes of the wavelet width learning system according to the gesture number defined in the step one and the feature number in the step five, and inputting the feature vectors obtained in the step five into the wavelet width learning system for classification; updating parameters according to the labeled samples in the training process; in the testing and practical process, judging the sequence number of the node with the highest excitation degree from a plurality of nodes output by the obtained wavelet width learning system, wherein the sequence number of the node is the sequence number of the classification action;
and step seven, according to the sequence number corresponding to each gesture defined in the step one, the type of the gesture can be identified through the sequence number of the classification result.
Preferably, the moving window method in the fourth step specifically includes:
first, signal data is extracted for a short period of time and square integration is performed, as shown in the following formula:
where S (t) is electromyographic signal data within the window, S i Representing t i Integration of the time signal, i.e. the energy value at the corresponding time; setting a threshold value beta to the energy value S i Make a judgment when t i Time of day energy value S i Greater than threshold beta and moving consecutive n after the window 1 The secondary energy values are all greater than the threshold value, the time t is considered i Is the starting time of the action; herein the base isOn the basis, when from t j Time of day energy value S j Less than threshold beta and having a succession n 2 The secondary energy value is less than the threshold value and is regarded as t j The action signal segment is therefore:
where k is the action detected in the kth segment. .
Preferably, the medium-small wave width learning system in the step six is specifically:
(1) Mapping different sets of feature nodes using the input vectors:
wherein w is i ,a i And b i The mapping weight, the transfer parameter and the scaling parameter are respectively generated randomly in the initialization process and updated by using a k-mean algorithm; phi (phi) i (. Cndot.) is the ith wavelet basis function, where n is the total number of wavelet basis functions;
(2) Series connection of each group of characteristic nodes:
Z n =[Z 1 ,Z 2 ,......,Z n ];
(3) Utilizing characteristic node Z n Mapping incremental nodes:
H m =ξ(Z n W h +β h ),
wherein W is h And beta h The weight and the threshold value parameters of the increment node are respectively generated randomly during initialization, so that the parameters are fixed and do not need to be updated; ζ (·) is a general excitation function, where a sigmoid function can be used;
(4) In the training process, the output weight parameters are obtained by using a pseudo-inverse value and a ridge regression algorithm:
W all =[Z n |H m ] + Y,
wherein Y is the parameter of the training setOutput of examination, while [ Z ] n |H m ] + Is [ Z n |H m ]Is a pseudo-inverse of (2); in the testing and practical process, the output node can be directly mapped
Compared with the prior art, the invention has the following beneficial effects:
compared with the algorithm based on the traditional deep learning network, the method can complete training of the model and determination of parameters more quickly, so that the working efficiency is improved; nodes can be dynamically expanded to increase the recognition rate of the system without the need to fully re-build and train the model.
Drawings
FIG. 1 is a schematic flow chart of the present invention;
FIG. 2 is a schematic diagram of a wavelet width learning system according to the present invention;
fig. 3 is an myoelectric signal acquisition device (Myo-arm ring) of the present invention.
Detailed Description
The present invention will be described in further detail with reference to examples and drawings, but embodiments of the present invention are not limited thereto.
As shown in fig. 1 to 3, a sEMG gesture recognition method based on a wavelet width learning system includes the following steps:
step one, defining the number d of gesture action types according to corresponding recognition scenes, and distributing a serial number for each gesture.
Step two, an electromyographic signal acquisition device shown in fig. 3 is used for acquiring an sEMG signal s.
Step three, filtering and denoising the sEMG original signal according to the frequency characteristic of the sEMG signal, and removing noise outside the frequency band of 10 Hz-500 Hz by using a Butterworth filter at the same time:
where N is the order, ω of the filter c Is a cut-off frequency filter.
Detecting an active segment in the sEMG signal by adopting a moving window method;
the moving window method specifically comprises the following steps:
first, signal data is extracted for a short period of time and square integration is performed, as shown in the following formula:
where S (t) is electromyographic signal data within the window, S i Representing t i Integration of the time signal, i.e. the energy value at the corresponding time; setting a threshold value beta to the energy value S i Make a judgment when t i Time of day energy value S i Greater than threshold beta and moving consecutive n after the window 1 The secondary energy values are all greater than the threshold value, the time t is considered i Is the starting time of the action; on the basis, when from t j Time of day energy value S j Less than threshold beta and having a succession n 2 The secondary energy value is less than the threshold value and is regarded as t j The action signal segment is therefore:
where k is the action detected in the kth segment.
Fifthly, detecting the detected movable section signal sectionExtracting features; calculating a characteristic of the signal segment, the characteristic comprising: average absolute value q 1 Root mean square q 2 Median frequency q 3 Mean frequency q 4 … … each feature of the tandem active segment is a feature vector:
x k =[q 1 ,q 2 ,...,q n ] T ,
wherein k is the kth active signal segment;
further, the feature vector of each active segment is taken as an input vector x of the wavelet width learning classification system.
Step six, designing input and output nodes of the wavelet width learning system according to the gesture number defined in the step one and the feature number in the step five, and inputting the feature vectors obtained in the step five into the wavelet width learning system for classification; updating parameters according to the labeled samples in the training process; in the testing and practical process, judging the sequence number of the node with the highest excitation degree from a plurality of nodes output by the obtained wavelet width learning system, wherein the sequence number of the node is the sequence number of the classification action;
the wavelet width learning system specifically comprises:
(1) Mapping different sets of feature nodes using the input vectors:
wherein w is i ,a i And b i The mapping weight, the transfer parameter and the scaling parameter are respectively generated randomly in the initialization process and updated by using a k-mean algorithm; phi (phi) i (. Cndot.) is the ith wavelet basis function, where n is the total number of wavelet basis functions;
(2) Series connection of each group of characteristic nodes:
Z n =[Z 1 ,Z 2 ,......,Z n ];
(3) Utilizing characteristic node Z n Mapping incremental nodes:
H m =ξ(Z n W h +β h ),
wherein W is h And beta h The weight and the threshold value parameters of the increment node are respectively generated randomly during initialization, so that the parameters are fixed and do not need to be updated; ζ (·) is a general excitation function, where a sigmoid function can be used;
(4) In the training process, the output weight parameters are obtained by using a pseudo-inverse value and a ridge regression algorithm:
W all =[Z n |H m ] + Y,
where Y is the reference output of the training set, and Z n |H m ] + Is [ Z n |H m ]Is a pseudo-inverse of (2); in the testing and practical process, the output node can be directly mapped
And step seven, according to the sequence number corresponding to each gesture defined in the step one, the type of the gesture can be identified through the sequence number of the classification result.
The method applies the wavelet width learning system to training and classifying in the gesture classification recognition algorithm, and has the advantages of combining the width learning system, being faster and more efficient compared with a deep learning network; the wavelet basis function is used as the excitation function of the feature layer, which results in an improved nonlinear fitting capability of the network.
Compared with the algorithm based on the traditional deep learning network, the method can complete training of the model and determination of parameters more quickly, so that the working efficiency is improved; nodes can be dynamically expanded to increase the recognition rate of the system without the need to fully re-build and train the model.
The foregoing is illustrative of the present invention and is not to be construed as limiting thereof, but rather as various changes, modifications, substitutions, combinations, and simplifications which may be made therein without departing from the spirit and principles of the invention are intended to be included within the scope of the invention.
Claims (2)
1. The sEMG gesture recognition method based on the wavelet width learning system is characterized by comprising the following steps of:
step one, defining the number d of gesture action types according to corresponding recognition scenes, and distributing a serial number to each gesture;
step two, an electromyographic signal acquisition device is used for acquiring sEMG signals s;
step three, filtering and denoising the sEMG original signal according to the frequency characteristic of the sEMG signal, and removing noise outside the frequency band of 10 Hz-500 Hz by using a Butterworth filter at the same time:
where N is the order, ω of the filter c Filtering for a cut-off frequency;
detecting an active segment in the sEMG signal by adopting a moving window method;
step five, extracting the characteristics of the detected movable section; calculating a characteristic of the signal segment, the characteristic comprising: average absolute value q 1 Root mean square q 2 Median frequency q 3 Mean frequency q 4 ,……,q n The individual features of the tandem active segment are feature vectors:
x k =[q 1 ,q 2 ,...,q n ] T ,
wherein k is the kth active signal segment;
further, taking the characteristic vector of each active segment as an input vector x of the wavelet width learning classification system;
step six, designing input and output nodes of the wavelet width learning system according to the gesture number defined in the step one and the feature number in the step five, and inputting the feature vectors obtained in the step five into the wavelet width learning system for classification; updating parameters according to the labeled samples in the training process; in the testing and practical process, judging the sequence number of the node with the highest excitation degree from a plurality of nodes output by the obtained wavelet width learning system, wherein the sequence number of the node is the sequence number of the classification action;
step seven, according to the serial number corresponding to each gesture defined in the step one, the type of the gesture can be identified through the serial number of the classification result;
the medium-small wave width learning system in the step six specifically comprises the following steps:
(1) Mapping different sets of feature nodes using the input vectors:
wherein w is i ,a i And b i The mapping weight, the transfer parameter and the scaling parameter are respectively generated randomly in the initialization process and updated by using a k-mean algorithm; phi (phi) i (. Cndot.) is the ith wavelet basis function, where n is the total number of wavelet basis functions;
(2) Series connection of each group of characteristic nodes:
Z n =[Z 1 ,Z 2 ,......,Z n ];
(3) Utilizing characteristic node Z n Mapping incremental nodes:
H m =ξ(Z n W h +β h ),
wherein W is h And beta h The weight and the threshold value parameters of the increment node are respectively generated randomly during initialization, so that the parameters are fixed and do not need to be updated; ζ (·) is a general excitation function, where a sigmoid function can be used;
(4) In the training process, the output weight parameters are obtained by using a pseudo-inverse value and a ridge regression algorithm:
W all =[Z n |H m ] + Y,
2. The sEMG gesture recognition method based on the wavelet width learning system according to claim 1, wherein the moving window method in the fourth step specifically comprises:
first, signal data is extracted for a short period of time and square integration is performed, as shown in the following formula:
where S (t) is electromyographic signal data within the window, S i Representing t i Integration of the time signal, i.e. the energy value at the corresponding time; setting a threshold value beta to the energy value S i Make a judgment when t i Time of day energy value S i Greater than threshold beta and moving consecutive n after the window 1 The secondary energy values are all greater than the threshold value, the time t is considered i Is the starting time of the action; on the basis, when from t j Time of day energy value S j Less than threshold beta and having a succession n 2 The secondary energy value is less than the threshold value and is regarded as t j The action signal segment is therefore:
where k is the action detected in the kth segment.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910548939.4A CN110389663B (en) | 2019-06-24 | 2019-06-24 | sEMG gesture recognition method based on wavelet width learning system |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910548939.4A CN110389663B (en) | 2019-06-24 | 2019-06-24 | sEMG gesture recognition method based on wavelet width learning system |
Publications (2)
Publication Number | Publication Date |
---|---|
CN110389663A CN110389663A (en) | 2019-10-29 |
CN110389663B true CN110389663B (en) | 2023-05-23 |
Family
ID=68285868
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910548939.4A Active CN110389663B (en) | 2019-06-24 | 2019-06-24 | sEMG gesture recognition method based on wavelet width learning system |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110389663B (en) |
Families Citing this family (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110826625B (en) * | 2019-11-06 | 2022-04-12 | 南昌大学 | Finger gesture classification method based on surface electromyographic signals |
CN111160392A (en) * | 2019-12-03 | 2020-05-15 | 广东工业大学 | Hyperspectral classification method based on wavelet width learning system |
CN111695446B (en) * | 2020-05-26 | 2021-07-27 | 浙江工业大学 | Gesture recognition method integrating sEMG and AUS |
CN113657479B (en) * | 2021-08-12 | 2022-12-06 | 广东省人民医院 | Novel multi-scale depth-width combined pathological picture classification method, system and medium |
CN114098768B (en) * | 2021-11-25 | 2024-05-03 | 哈尔滨工业大学 | Cross-individual surface electromyographic signal gesture recognition method based on dynamic threshold and EasyTL |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106236336A (en) * | 2016-08-15 | 2016-12-21 | 中国科学院重庆绿色智能技术研究院 | A kind of myoelectric limb gesture and dynamics control method |
CN109062401A (en) * | 2018-07-11 | 2018-12-21 | 北京理工大学 | A kind of real-time gesture identifying system based on electromyography signal |
CN109077715A (en) * | 2018-09-03 | 2018-12-25 | 北京工业大学 | A kind of electrocardiosignal automatic classification method based on single lead |
CN109308521A (en) * | 2018-08-27 | 2019-02-05 | 广东工业大学 | A kind of quaternary SerComm degree study filtering method for eliminating Physiological tremor |
Family Cites Families (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
KR102619981B1 (en) * | 2016-02-02 | 2024-01-02 | 삼성전자주식회사 | Gesture classification apparatus and method using electromyogram signals |
-
2019
- 2019-06-24 CN CN201910548939.4A patent/CN110389663B/en active Active
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106236336A (en) * | 2016-08-15 | 2016-12-21 | 中国科学院重庆绿色智能技术研究院 | A kind of myoelectric limb gesture and dynamics control method |
CN109062401A (en) * | 2018-07-11 | 2018-12-21 | 北京理工大学 | A kind of real-time gesture identifying system based on electromyography signal |
CN109308521A (en) * | 2018-08-27 | 2019-02-05 | 广东工业大学 | A kind of quaternary SerComm degree study filtering method for eliminating Physiological tremor |
CN109077715A (en) * | 2018-09-03 | 2018-12-25 | 北京工业大学 | A kind of electrocardiosignal automatic classification method based on single lead |
Non-Patent Citations (1)
Title |
---|
基于小波特征分析的手指动作识别研究;李博等;《现代生物医学进展》;20111030(第20期);第3942-3945页 * |
Also Published As
Publication number | Publication date |
---|---|
CN110389663A (en) | 2019-10-29 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110389663B (en) | sEMG gesture recognition method based on wavelet width learning system | |
CN113128552B (en) | Electroencephalogram emotion recognition method based on depth separable causal graph convolution network | |
CN109859147A (en) | A kind of true picture denoising method based on generation confrontation network noise modeling | |
CN103366180A (en) | Cell image segmentation method based on automatic feature learning | |
CN102169690A (en) | Voice signal recognition system and method based on surface myoelectric signal | |
CN101859377A (en) | Electromyographic signal classification method based on multi-kernel support vector machine | |
CN110135365B (en) | Robust target tracking method based on illusion countermeasure network | |
JP2018511870A (en) | Big data processing method for segment-based two-stage deep learning model | |
CN103971329A (en) | Cellular nerve network with genetic algorithm (GACNN)-based multisource image fusion method | |
CN108205727B (en) | Digital workshop process decision method based on decision tree and expert system | |
CN112541415B (en) | Brain muscle function network motion fatigue detection method based on symbol transfer entropy and graph theory | |
CN109086823B (en) | Automatic statistical method for wheat scab ear disease rate | |
CN111603158B (en) | Fatigue driving warning method and system based on electrophysiological signal artificial intelligent analysis | |
CN112651426A (en) | Fault diagnosis method for rolling bearing of wind turbine generator | |
CN113116361A (en) | Sleep staging method based on single-lead electroencephalogram | |
CN106097257A (en) | A kind of image de-noising method and device | |
CN108921106A (en) | A kind of face identification method based on capsule | |
CN114581470B (en) | Image edge detection method based on plant community behaviors | |
CN115859055A (en) | Feature extraction method for multi-source heterogeneous big data in aircraft manufacturing process | |
CN104021563B (en) | Method for segmenting noise image based on multi-objective fuzzy clustering and opposing learning | |
CN113876339A (en) | Method for constructing sleep state electroencephalogram characteristic signal feature set | |
Andronache et al. | Automatic gesture recognition framework based on forearm emg activity | |
CN115581467A (en) | SSVEP (steady state visual evoked potential) identification method based on time, frequency and time-frequency domain analysis and deep learning | |
CN102327116B (en) | Network system random resonance restoration method of cortex electroencephalographic signal | |
CN110399656B (en) | Lower-loading waist-saving parameter design method based on fuzzy logic and neural network |
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 |