Background technology
Along with the develop rapidly of microelectric technique, sensor technology and computer technology, handheld mobile device, Wearable and microcomputer are universal in people's daily life gradually.Such as, but due to the restriction of use scenes and mini-plant, traditional human-computer interaction device, the equipment such as keyboard, mouse can not meet the demand of people.Transportable miniaturization gesture identification equipment is suggested as novel human-computer interaction device.
Muscle electric signal, since 1945 are used in control field, experienced by the development of last 100 years, is always studiedly applied to medical diagnosis and bio-mechanical field.Along with the development of biomedical technology, artificial intelligence technology, the method using electromyographic signal to carry out gesture identification is suggested and constantly explores.Wherein, EMG signal real-time processing and identify it is an important difficult point.
Existing electromyographic signal gesture identification method does not generally break away from the thinking of traditional algorithm, the algorithm for pattern recognition that end user's artificial neural networks (ANN), support vector machine (SVM) etc. are comparatively complicated.And adopt the frequency domain character of extraction signal or time-frequency characteristics to input as algorithm.
The identification using classic method to carry out electromyographic signal can reach good effect, but higher algorithm complex and storage space take the requirements such as low-power consumption, low computational complexity, in real time process, the data structure that cannot meet embedded system are simple.Therefore, in the urgent need to a kind of Method of Surface EMG Pattern Recognition that can be applied to embedded system.
Summary of the invention
In view of this, be necessary to provide a kind of Method of Surface EMG Pattern Recognition.
The invention provides a kind of Method of Surface EMG Pattern Recognition, the method comprises the steps: that a. samples the electromyographic signal of different passage; B. analog to digital conversion is carried out to the electromyographic signal of described sampling, feature extraction is carried out to the electromyographic signal after analog to digital conversion; C. proper vector integration is carried out according to the section feature of said extracted; Enter Modling model process or model does not exist if be d. set up, then utilize linear regression analysis method establishment model; If e. model exists, then the proper vector of above-mentioned integration and the model of foundation is utilized to carry out gesture identification.
Preferably, the method also comprises step between step c and steps d: judge whether to need Modling model.
Preferably, described step b specifically comprises: in each section, L data are as the feature extraction of the data source section of carrying out.
Preferably, the section feature of described extraction comprises: the conversion of zero crossing number, absolute average, slope and waveform length.
Preferably, described step c specifically comprises: combined by channel sequence in the section feature that section is extracted at the same time by M passage, and generate integration characteristics vector, vector dimension is [M × 3,1].
Preferably, utilize linear regression analysis method establishment model specifically to comprise in described steps d: in model training process, the feature extracted during different gesture is carried out classification annotation by user.Linear regression analysis method establishment model is utilized to the multiple proper vectors marked.
Preferably, utilize the model of the proper vector of above-mentioned integration and foundation to carry out gesture identification in described step e specifically to comprise: the arm muscles electric signal obtained by real-time sampling generates integration characteristics vector according to the method for step a to step c, integration characteristics vector is carried out computing with the model set up compare, calculate corresponding classified information, and then identification obtains real-time gesture information.
The invention provides a kind of Method of Surface EMG Pattern Recognition, different finger gestures can be gone out by discriminator, and when basic guarantee discrimination, the computational complexity that traditional method is brought be reduced greatly.Performance be 1.25DMIPS/MHz, dominant frequency is on the microcontroller of 84MHz, training process computing required time is less than 30ms, and Real time identification computing required time is less than 10ms.Meanwhile, the discrimination of more than 90% is obtained when recognition category number C is less than 10.Efficiency significantly improves, and can meet the requirement of human-computer interaction device for real-time simultaneously.
Embodiment
Below in conjunction with drawings and the specific embodiments, the present invention is further detailed explanation.
Consulting shown in Fig. 1, is the operation process chart of Method of Surface EMG Pattern Recognition preferred embodiment of the present invention.
Step S1, the electromyographic signal of different passage of sampling.Specifically:
The arm muscles electric signal of different passage is obtained according to fixing sample frequency, i.e. electromyographic signal, described electromyographic signal is stored in embedded microcontroller (μ C) or microprocessor (μ P), and is that L carries out homogenous segmentations according to segment length.
Step S2, carries out analog to digital conversion to the electromyographic signal of described sampling, carries out feature extraction to the electromyographic signal after analog to digital conversion.Specifically:
In each section, L data are as the feature extraction of the data source section of carrying out, and the section feature of described extraction comprises: the conversion of zero crossing number, absolute average, slope and waveform length.Computing formula is as follows:
Zero crossing number:
Suppose there are two sample x
kand x
k+1if met: { x
k>0 and x
k+1<0} or { x
k<0 and x
k+1>0}, then zero crossing number adds one.Wherein x
kit is a kth sample.
Absolute average:
wherein, x
kbe a kth sample, L is number of samples.
Slope converts:
Suppose there are two sample x
kand x
k+1if met: { x
k>x
k-1and x
k>x
k+1or { x
k<x
k-1and x
k<x
k+1, then slope conversion number adds one, wherein x
kit is a kth sample.
Waveform length:
wherein, △ x
k=x
k-x
k-1, x
kit is a kth sample.
Step S3, the section feature according to said extracted carries out proper vector integration.Specifically:
Combined by channel sequence in the section feature that section is extracted at the same time by M passage, generate integration characteristics vector, vector dimension is [M × 3,1].
Step S4, judges whether to need Modling model: enter Modling model process or model does not exist if be set up, then enter step S5 Modling model; If model exists, then enter step S6.
Step S5, utilizes linear regression analysis method establishment model.Specifically:
In model training process, the feature extracted during different gesture is carried out classification annotation by user.Linear regression analysis method establishment model is utilized to the multiple proper vectors marked.That is to say:
In model training process, use the method for linear regression analysis that multiple sections of features are carried out model training as training sample, final generation can identify the disaggregated model of C class integration characteristics, respectively a corresponding C gesture.
The model set up comprises multiple projection vector and multiple border.
Step S6, utilizes the proper vector of above-mentioned integration and the model of foundation, carries out gesture identification.Specifically:
The arm muscles electric signal obtained by real-time sampling, according to the method generation integration characteristics vector of step S1 to S3, integration characteristics vector is carried out computing with the model set up and compares, calculate corresponding classified information, and then identification obtains real-time gesture information.
In identifying, user makes the defined and gesture marked, and the electromyographic signal feature of gesture is extracted, and calculates, obtain final recognition result after above-mentioned process with the model set up.Described recognition result corresponds to the gesture-type of user's mark when training.
It should be noted that, after model has been set up, enter gesture identification.The starting point of gesture identification is the starting point of whole method, return step S1 electromyographic signal sampling element, repeat electromyographic signal sampling, electromyographic signal feature extraction, proper vector integration (being also step S1 to S3), and coordinate the model set up, export recognition result.Continuous circulation identifies, exports gesture identification result.
Although the present invention is described with reference to current better embodiment; but those skilled in the art will be understood that; above-mentioned better embodiment is only used for the present invention is described; not be used for limiting protection scope of the present invention; any within the spirit and principles in the present invention scope; any modification of doing, equivalence replacement, improvement etc., all should be included within the scope of the present invention.