CN110598676B - Deep learning gesture electromyographic signal identification method based on confidence score model - Google Patents
Deep learning gesture electromyographic signal identification method based on confidence score model Download PDFInfo
- Publication number
- CN110598676B CN110598676B CN201910912473.1A CN201910912473A CN110598676B CN 110598676 B CN110598676 B CN 110598676B CN 201910912473 A CN201910912473 A CN 201910912473A CN 110598676 B CN110598676 B CN 110598676B
- Authority
- CN
- China
- Prior art keywords
- data
- model
- electromyographic
- confidence score
- gesture
- 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
- 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
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/02—Preprocessing
-
- 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
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/12—Classification; Matching
Abstract
The deep learning gesture electromyographic signal identification method based on the confidence score model comprises the following steps of: the collected myoelectricity original data are preprocessed; after preprocessing, respectively labeling categories by using labels, and then segmenting each category into a plurality of segments of data through a sliding window; dividing the data into training data and testing data, and respectively randomly sequencing the training data and the testing data; normalizing all the data; converting the frequency spectrum data into discrete Fourier transform, calculating the energy of each frequency according to the frequency spectrum, and using the energy of different frequencies to form a sequence as the input of a classification model; 4 classification models are formed by combining 4 sets of original electromyographic data and spectrum data by using two classification methods of a convolutional neural network model and a residual error network model, and the models are respectively and independently trained; and acquiring the recognition result of each model to the gesture, combining the results of different models by using the confidence score, and taking the final result with the highest total confidence score.
Description
Technical Field
The invention belongs to the field of biological signal processing, and particularly relates to a deep learning gesture electromyographic signal identification method based on a confidence score model.
Background
Electromyographic signals are a superposition of Motor Unit Action Potentials (MUAPs) in a multitude of muscle fibers, both temporally and spatially. The electromyographic signals recorded by the electromyograph can be divided into the following types according to different recording modes: surface electromyography signals and needle electrode electromyography signals, both of which contain the anatomical and physiological properties of the muscle. Surface Electromyography (sEMG) is a comprehensive effect of superficial muscle Electromyography and electrical activity of a nerve trunk on the surface of the skin, and can reflect the activity of neuromuscular to a certain extent. Compared with the needle electrode electromyographic signals, the surface electromyographic signals have the advantages of non-invasiveness, no wound, simplicity in operation and the like in measurement. Therefore, the surface electromyographic signals have important application values in the aspects of clinical medicine, rehabilitation medicine, human-computer interfaces and the like.
Gesture recognition based on surface electromyographic signals is an important research field of human-computer interfaces. The conventional myoelectric gesture recognition process can be divided into the following steps. First, the mathematical methods needed to artificially design features and extract features. Common artificial features include root mean square and average value, and a spectrum feature is also a feature frequently adopted in the conventional method, and represents some global characteristics of signals in a frequency domain, and observation of the spectrum feature is easy to find that different gestures have significant differences in various frequency bands. The different features are then combined to form a feature set, which is typically represented by a vector. Next, the high-dimensional vector is subjected to dimensionality reduction by a method such as principal component analysis. And finally, mapping the vectors containing the features to different categories by using a machine learning method to achieve the purpose of gesture classification.
The traditional recognition method has the defects of low classification accuracy and limited application scenes under the condition of more gesture categories. In recent years, deep learning methods have begun to be applied to the recognition of myoelectric gesture signals. Research shows that the electromyographic signals are automatically extracted by a deep learning method, the characteristic effect is better than that of the traditional manual design, and the classification task with more categories is remarkably improved. In the application scenario of gesture recognition, a cyclic neural network (RNN) and a Convolutional Neural Network (CNN) are commonly used as deep learning methods. The cyclic neural network has a limited ability to recognize longer sequences, particularly electromyographic signals with the sequence length of hundreds of orders of magnitude, and later data in time has a greater influence on the cyclic neural network than earlier data, so that the early data has a smaller weight on the recognition result. The convolutional neural network can find the relevant characteristics of the whole sequence, and the longer sequence length can enable the identification to be more accurate, so that the convolutional neural network is suitable for electromyographic classification tasks. Because the automatic feature extraction process of the deep learning method is invisible, the deviation of the model cannot be corrected manually, and a single model often has a high recognition error rate.
Disclosure of Invention
Aiming at the defects of a single model, the invention provides a myoelectricity recognition algorithm adopting multi-model combination. After spectral and filtering processing is performed on the raw data, features are extracted using a deep neural network model. Studies have shown that there are models of feature differences, the combination of which can effectively reduce the overall error rate. Therefore, the method trains a differential model by combining different model structures with different data sets, and then integrates the recognition results of a plurality of models through confidence score so as to improve the recognition rate.
The deep learning gesture electromyographic signal identification method based on the confidence score model comprises the following steps of:
step 1: acquiring electromyographic data; acquiring original myoelectric signals by adopting electrode equipment, and preprocessing the acquired myoelectric original data, including band-pass filtering and abnormal point removal; respectively labeling the preprocessed electromyographic signals with a label, and then dividing each category into a plurality of segments of data through a sliding window; then, dividing the data into training data and testing data, and respectively randomly sequencing the training data and the testing data; finally, normalizing all the data;
step 2: carrying out frequency spectrum transformation; the preprocessed electromyographic data is converted into frequency spectrum data by using discrete Fourier transform, a Hamming window is required to be smooth in the frequency spectrum conversion process, the energy of each frequency is calculated according to the frequency spectrum, and different frequency energies form a sequence to be used as the input of a classification model;
and step 3: training a model; 4 classification models are formed by combining 4 sets of original electromyographic data and spectrum data by using two classification methods of a convolutional neural network model and a residual error network model, and the models are respectively and independently trained; the training process uses a cross-validation method to prevent overfitting;
and 4, step 4: combining the models; and acquiring the recognition result of each model to the gesture, combining the results of different models by using the confidence score, and taking the final result with the highest total confidence score.
Further, in step 1, 12 electrode devices for acquiring the original electromyographic signals are patch electrodes, the electrodes are connected to the signal acquisition device through leads, the patch electrodes are located at the arm or the wrist, and the sampling frequency adopts a high-frequency sampling mode higher than 2000 Hz.
Further, in step 1, in the sliding window division, specifically, the signal is divided into signals with the size of 100ms to 300ms, and there is a sliding step with the length of 20ms, and the divided signals are not more than 300 ms.
Further, in step 2, the spectrum transformation is performed by discrete Fourier transform, as shown in formula (1), and for the N point sequence { x [ N ] }, 0 ≦ N < N, there are:
where e is the base of the natural logarithm and i is the unit of an imaginary number;
and calculating the energy value of each frequency after the frequency spectrum transformation to obtain an energy spectrogram, and taking the numerical value of the frequency range in which the corresponding electromyographic signal numerical value is concentrated in the energy spectrogram as a characteristic point.
Further, in step 3, the method for recognizing deep learning gesture electromyographic signals based on the confidence score model according to claim 1 is characterized in that: in step 3, the cost function used by the training model is cross entropy, as shown in formula (2):
wherein C represents the cost, x represents the sample, y represents the actual value, a represents the output value, and n represents the total number of samples;
the value of the cost function is the basis of model learning, the model is learned by using a back propagation method according to the cost, and the back propagation is as shown in (3):
wherein the content of the first and second substances,
represents the output of the jth neuron at the l-th level,
represents the weight of the kth neuron connected to the jth neuron of the l-1 layer
Represents the bias of the jth neuron of the ith layer, and σ represents the activation function.
The gesture corresponding to the electromyographic data can be predicted by using the model obtained by cross entropy cost and back propagation method training.
Further, in step 4, the confidence score method is to add the results of all classifiers by a weight average, and the final result is determined by the score of each gesture, and the specific steps are as follows: step 4-1, in a classification task, predicting a single electromyographic signal by using a model, and generating a gesture and a corresponding confidence score; assuming a total of N electromyographic signals, N gestures and corresponding confidence scores will be generated;
and 4-2, assuming that a total of M models participate in the classification task, for a single electromyographic signal, M gestures and confidence scores are obtained, and the confidence scores identical to the gestures are added. The gesture with the highest confidence score is used as the final gesture of the single electromyographic signal;
and 4-3, repeating the method of the step 4-2 on the N electromyographic signals of the classification task to obtain the final gestures of all the N electromyographic signals.
The invention has the following beneficial results:
(1) by integrating the characteristics and results of a plurality of models, the deviation of characteristic extraction existing in a single model is overcome, and the error rate of identification can be effectively reduced.
(2) The traditional model combination adopts a method of data set diversity to further increase the difference between models and the diversity of the models through the differentiation of different constituent models of a classification method.
(3) The invention adopts the residual error network model as one of the combination models, and the residual error network can limit the overfitting degree of the model and effectively inhibit the negative influence caused by the over-deep depth of the model.
Drawings
Fig. 1 is a flow chart of converting raw data into spectral data.
Fig. 2 is an explanatory diagram of the deep learning confidence score model architecture.
Detailed Description
The technical scheme of the invention is further explained in detail by combining the drawings in the specification.
The deep learning gesture electromyographic signal identification method based on the confidence score model is characterized by comprising the following steps of: comprises the following steps:
step 1: acquiring electromyographic data; acquiring an original myoelectric signal by adopting electrode equipment, and preprocessing the acquired myoelectric original data, including band-pass filtering and abnormal point removal; respectively labeling the preprocessed electromyographic signals with a label, and then dividing each category into a plurality of segments of data through a sliding window; then, dividing the data into training data and testing data, and respectively randomly sequencing the training data and the testing data; and finally, normalizing all the data.
Step 2: carrying out frequency spectrum transformation; the preprocessed electromyographic data is converted into frequency spectrum data by using discrete Fourier transform, a Hamming window is required to be smooth in the frequency spectrum conversion process, the energy of each frequency is calculated according to the frequency spectrum, and different frequency energies form a sequence to be used as the input of a classification model.
And step 3: training a model; 4 classification models are formed by combining 4 sets of original electromyographic data and spectrum data by using two classification methods of a convolutional neural network model and a residual error network model, and the models are respectively and independently trained; the training process uses a cross-validation method to prevent overfitting.
And 4, step 4: combining the models; and acquiring the recognition result of each model to the gesture, combining the results of different models by using the confidence score, and taking the final result with the highest total confidence score.
In the step 1, 12 electrode devices for acquiring the original electromyographic signals are provided, the electrodes are connected to a signal acquisition device through leads, the patch electrodes are positioned at the arm or the wrist, and the sampling frequency adopts a high-frequency sampling mode higher than 2000 Hz.
In step 1, in the sliding window division, signals are divided into signals with the size of 100ms to 300ms, a sliding step with the length of 20ms is provided, and the divided signals are not more than 300 ms. Since the time taken to acquire a signal in practical applications determines the length of the signal and also determines the delay of the system. The deep learning method is different from the traditional classification method, a large amount of data are needed for training a model, overfitting is often caused by insufficient data, and the prediction accuracy is low. Therefore, a smaller sliding step can obtain a large amount of data, so that the depth model can be sufficiently trained.
Next, normalization operation is performed on the entire signal, and the benefit of normalization operation is that the gradient descent direction can be further biased toward the optimal solution direction, and the convergence rate of the model can be improved. And secondly, the accuracy of the model is improved, because the electromyographic signals of partial channels have smaller orders of magnitude compared with those of other channels. The normalization operation reduces the influence of the data magnitude difference on the model and ensures the reliability of the result.
In step 2, the whole spectrum transformation process is shown in fig. 1. The spectral transformation is performed by discrete Fourier transform, as shown in equation (1), for a sequence of N points { x [ N ] }, 0 ≦ N < N, having:
wherein e is the base of the natural logarithm and i is the imaginary unit;
the time domain signal is projected to the frequency domain by the frequency spectrum transformation, so that the integral characteristic of the electromyographic signal can be analyzed. And calculating the energy value of each frequency after the frequency spectrum transformation to obtain an energy spectrogram, and taking the numerical value of the frequency range in which the corresponding electromyographic signal numerical value is concentrated in the energy spectrogram as a characteristic point.
In step 3, the method adopts two models of a convolutional neural network and a residual error network. Convolutional neural networks are a representation learning model that is able to automatically discover features for classification by inputting raw data into the model. The residual error network is an improvement of the convolutional neural network, and introduces the residual error connection operation on the basis of the original model, so that the training speed of the model is higher, and the problem of gradient disappearance easily occurring in a deep convolutional network is solved.
Research shows that when the model is trained by using different coding modes of the same data set, the extracted features of the model can be different. The spectral data is obtained by linear transformation of time series data, and thus can be regarded as another encoding mode of the time series data. Therefore, two models with the same structure are trained by using time sequence data and spectrum data respectively, and the models have differences in extracted features, so that the classification results have differences.
The cost function used by the training model is the cross entropy, as shown in equation (2):
wherein C represents the cost, x represents the sample, y represents the actual value, a represents the output value, and n represents the total number of samples;
the value of the cost function is the basis of model learning, the model is learned by using a back propagation method according to the cost, and the back propagation is as shown in (3):
wherein the content of the first and second substances,
represents the output of the jth neuron at the l-th level,
represents the weight of the kth neuron connected to the jth neuron of the l-1 layer
Represents the bias of the jth neuron of the ith layer, and σ represents the activation function.
The gesture corresponding to the electromyographic data can be predicted by using the model obtained by cross entropy cost and back propagation method training.
In step 4, assuming that there is a classification task and a series of models that can perform the classification task individually, and the classification errors generated by the models are not caused by the same reason, averaging the results of the classifiers will reduce the overall classification error rate. The confidence score method is to add the results of all classifiers by weight, and the final result is determined by the score of each gesture, and the specific steps are as follows: step 4-1, in a classification task, predicting a single electromyographic signal by using a model, and generating a gesture and a corresponding confidence score; assuming a total of N electromyographic signals, N gestures and corresponding confidence scores will be generated;
and 4-2, assuming that a total of M models participate in the classification task, for a single electromyographic signal, M gestures and confidence scores are obtained, and the confidence scores identical to the gestures are added. The gesture with the highest confidence score is used as the final gesture of the single electromyographic signal;
and 4-3, repeating the method of the step 4-2 on the N electromyographic signals of the classification task to obtain the final gestures of all the N electromyographic signals.
The model number M is 4 as an example. The data size N may be any size. The specific architecture is shown in fig. 2. The models are CNN _1, CNN _2, Res _1 and Res _2, respectively. Where CNN _1 and CNN _2 are convolutional neural networks of the same structure, and Res _1 and Res _2 are residual networks of the same structure. However, CNN _2 and Res _2 add spectral shifts, which are characteristically different from the original model. The 4 models will get the final result of the whole architecture using the confidence score method.
The confidence score has low calculation complexity, occupies less calculation resources, has high speed, occupies far lower calculation resources than a depth model, supports the parallel calculation of an M model, and can be applied to gesture classification tasks with high real-time performance.
The above description is only a preferred embodiment of the present invention, and the scope of the present invention is not limited to the above embodiment, but equivalent modifications or changes made by those skilled in the art according to the present disclosure should be included in the scope of the present invention as set forth in the appended claims.
Claims (5)
1. The deep learning gesture electromyographic signal identification method based on the confidence score model is characterized by comprising the following steps of: comprises the following steps:
step 1: acquiring electromyographic data; acquiring an original myoelectric signal by adopting electrode equipment, and preprocessing the acquired myoelectric original data, including band-pass filtering and abnormal point removal; respectively labeling the preprocessed electromyographic signals with a label, and then dividing each category into a plurality of segments of data through a sliding window; then, dividing the data into training data and testing data, and respectively randomly sequencing the training data and the testing data; finally, normalizing all the data;
step 2: carrying out frequency spectrum transformation; the preprocessed electromyographic data is converted into frequency spectrum data by using discrete Fourier transform, a Hamming window is required to be smooth in the frequency spectrum conversion process, the energy of each frequency is calculated according to the frequency spectrum, and different frequency energies form a sequence to be used as the input of a classification model;
in step 2, the spectral transformation is performed by discrete fourier transform, as shown in formula (1), for a sequence of N points { x [ N ] }, N is greater than or equal to 0 and less than N, there are:
where e is the base of the natural logarithm and i is the unit of an imaginary number;
calculating the energy value of each frequency after the frequency spectrum transformation to obtain an energy spectrogram, and taking the numerical value of the frequency range in which the corresponding electromyographic signal numerical value is concentrated in the energy spectrogram as a characteristic point;
and step 3: training a model; 4 classification models are formed by combining 4 sets of original electromyographic data and spectrum data by using two classification methods of a convolutional neural network model and a residual error network model, and the models are respectively and independently trained; the training process uses a cross-validation method to prevent overfitting;
and 4, step 4: combining the models; and acquiring the recognition result of each model to the gesture, combining the results of different models by using the confidence score, and taking the final result with the highest total confidence score.
2. The deep learning gesture electromyographic signal recognition method based on the confidence score model according to claim 1, wherein: in the step 1, 12 electrode devices for acquiring the original electromyographic signals are provided, the electrodes are connected to a signal acquisition device through leads, the patch electrodes are positioned at the arm or the wrist, and the sampling frequency adopts a high-frequency sampling mode higher than 2000 Hz.
3. The deep learning gesture electromyographic signal recognition method based on the confidence score model according to claim 1, wherein: in step 1, in the sliding window division, signals are divided into signals with the size of 100ms to 300ms, a sliding step with the length of 20ms is provided, and the divided signals are not more than 300 ms.
4. The deep learning gesture electromyographic signal recognition method based on the confidence score model according to claim 1, wherein: in step 3, the cost function used by the training model is cross entropy, as shown in formula (2):
wherein C represents the cost, x represents the sample, y represents the actual value, a represents the output value, and n represents the total number of samples;
the value of the cost function is the basis of model learning, the model is learned by using a back propagation method according to the cost, and the back propagation is as shown in (3):
wherein the content of the first and second substances,
represents the output of the jth neuron at the l-th level,
represents the weight of the kth neuron connected to the jth neuron of the l-1 layer
Represents the bias of the jth neuron of the ith layer, and σ represents the activation function;
the gesture corresponding to the electromyographic data can be predicted by using the model obtained by cross entropy cost and back propagation method training.
5. The deep learning gesture electromyographic signal recognition method based on the confidence score model according to claim 1, wherein: in step 4, the confidence score method is to add the results of all classifiers by weight, and the final result is determined by the score of each gesture, and the specific steps are as follows: step 4-1, in a classification task, predicting a single electromyographic signal by using a model, and generating a gesture and a corresponding confidence score; assuming a total of N electromyographic signals, N gestures and corresponding confidence scores will be generated;
step 4-2, assuming that a total of M models participate in the classification task, for a single electromyographic signal, M gestures and confidence scores are obtained, and the confidence scores which are the same as the gestures are added; the gesture with the highest confidence score is used as the final gesture of the single electromyographic signal;
and 4-3, repeating the method of the step 4-2 on the N electromyographic signals of the classification task to obtain the final gestures of all the N electromyographic signals.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910912473.1A CN110598676B (en) | 2019-09-25 | 2019-09-25 | Deep learning gesture electromyographic signal identification method based on confidence score model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910912473.1A CN110598676B (en) | 2019-09-25 | 2019-09-25 | Deep learning gesture electromyographic signal identification method based on confidence score model |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110598676A CN110598676A (en) | 2019-12-20 |
| CN110598676B true CN110598676B (en) | 2022-08-02 |
Family
ID=68863363
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910912473.1A Active CN110598676B (en) | 2019-09-25 | 2019-09-25 | Deep learning gesture electromyographic signal identification method based on confidence score model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110598676B (en) |
Families Citing this family (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112183751A (en) * | 2020-11-09 | 2021-01-05 | 大连东软教育科技集团有限公司 | Neural network model prediction confidence calibration method, system and storage medium |
| CN112364762B (en) * | 2020-11-10 | 2024-02-27 | 南京大学 | Mechanical transmission fault detection method based on step error frequency spectrum characteristics |
| CN114515146B (en) * | 2020-11-17 | 2024-03-22 | 北京机械设备研究所 | Intelligent gesture recognition method and system based on electrical measurement |
| CN113312994A (en) * | 2021-05-18 | 2021-08-27 | 中国科学院深圳先进技术研究院 | Gesture classification recognition method and application thereof |
| CN116662773A (en) * | 2022-03-29 | 2023-08-29 | 深圳市应和脑科学有限公司 | Model acquisition system, gesture recognition method, gesture recognition device, apparatus and storage medium |
| CN114931389A (en) * | 2022-04-27 | 2022-08-23 | 福州大学 | Electromyographic signal identification method based on residual error network and graph convolution network |
| CN114611563B (en) * | 2022-05-16 | 2022-08-23 | 北京大学第三医院（北京大学第三临床医学院） | Method for identifying neurogenic damage of pin-pole electromyogram across parts |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105446484A (en) * | 2015-11-19 | 2016-03-30 | 浙江大学 | Electromyographic signal gesture recognition method based on hidden markov model |
| 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 |
| 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 |
-
2019
- 2019-09-25 CN CN201910912473.1A patent/CN110598676B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105446484A (en) * | 2015-11-19 | 2016-03-30 | 浙江大学 | Electromyographic signal gesture recognition method based on hidden markov model |
| 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 |
| 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 |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110598676A (en) | 2019-12-20 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110598676B (en) | Deep learning gesture electromyographic signal identification method based on confidence score model | |
| CN107714023B (en) | Static electrocardiogram analysis method and device based on artificial intelligence self-learning | |
| CN107951485B (en) | Ambulatory ECG analysis method and apparatus based on artificial intelligence self study | |
| CN110338786B (en) | Epileptic discharge identification and classification method, system, device and medium | |
| CN104523266B (en) | A kind of electrocardiosignal automatic classification method | |
| CN110781945A (en) | Electroencephalogram signal emotion recognition method and system integrating multiple features | |
| CN112426160A (en) | Electrocardiosignal type identification method and device | |
| CN109009098B (en) | Electroencephalogram signal feature identification method under motor imagery state | |
| CN111783715B (en) | Identity recognition method based on pulse signal feature extraction | |
| CN114052744A (en) | Electrocardiosignal classification method based on pulse neural network | |
| CN114081494B (en) | Depression state detecting system based on brain lateral reins signal | |
| CN114532993A (en) | Automatic detection method for electroencephalogram high-frequency oscillation signals of epileptic | |
| CN114533089A (en) | Lower limb action recognition and classification method based on surface electromyographic signals | |
| Phinyomark et al. | Applications of variance fractal dimension: A survey | |
| CN114384999A (en) | User irrelevant myoelectricity gesture recognition system based on self-adaptive learning | |
| Kaur et al. | Multi-class support vector machine classifier in EMG diagnosis | |
| CN111242100B (en) | Action identification method based on GST and VL-MOBPNN | |
| CN116712099A (en) | Fetal heart state detection method based on multi-mode data, electronic equipment and storage medium | |
| CN116019467A (en) | Feature extraction method of facial nerve electromyographic signals | |
| CN114343679A (en) | Surface electromyogram signal upper limb action recognition method and system based on transfer learning | |
| CN114469041A (en) | Heart rate change data characteristic analysis method in exercise process | |
| CN113907770A (en) | Method and system for detecting and identifying spike-slow complex wave based on feature fusion | |
| Şengür et al. | Classification of amyotrophic lateral sclerosis and healthy electromyography signals based on transfer learning | |
| CN114343680B (en) | Real-time decomposition method for dynamic surface electromyographic signals | |
| Dai et al. | Experimental study on neuronal spike sorting methods |
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 |