阅读说明：本技术 一种基于神经网络的肌肉运动单元提取方法 (muscle movement unit extraction method based on neural network ) 是由 何金保 安鹏 胡庆波 骆再飞 于 2019-09-11 设计创作，主要内容包括：本发明提供了一种基于神经网络的肌肉运动单元提取方法,针对肌肉动态收缩过程中,通过阵列式表面肌电信号,采用神经网络提取肌肉运动单元。本发明采用ART2神经网络对肌肉运动单元波形进行分类,分类过程中不需要进行神经网络训练,直接输入实际波形信号进行识别,简化了识别过程,节约运行时间。在分类过程中,为了应对肌肉动态收缩时运动单元的变化,动态更新分类模板。本发明在确认肌肉运动单元时,考虑了波形传播特性,提高了提取肌肉运动单元的准确性。本发明实现简单,满足实际应用的需要。(the invention provides a muscle movement unit extraction method based on a neural network, which is used for extracting a muscle movement unit by adopting the neural network through array type surface electromyographic signals in the dynamic muscle contraction process. According to the invention, the ART2 neural network is adopted to classify the waveform of the muscle movement unit, the neural network training is not needed in the classification process, the actual waveform signal is directly input for identification, the identification process is simplified, and the operation time is saved. In the classification process, the classification template is dynamically updated in order to deal with the change of the motion unit when the muscle is dynamically contracted. When the muscle movement unit is confirmed, the waveform propagation characteristic is considered, and the accuracy of extracting the muscle movement unit is improved. The invention is simple to realize and meets the requirement of practical application.)

1. A muscle movement unit extraction method based on a neural network is characterized by comprising the following steps:

Collecting array type surface electromyographic signals when muscles contract dynamically, filtering the signals and weakening interference;

Step two, extracting the muscle movement unit on each channel to send out waveforms;

Inputting the waveform emitted by the muscle movement unit of each channel into an ART2 neural network, and setting a nonlinear function in a neuron to suppress noise;

step four, in the ART2 recognition and classification process, dynamically updating the classification template, comparing the distribution waveform similarity and outputting the classification result;

Step five, carrying out differential operation on the classification results output by each channel, and confirming the muscle movement unit according to the propagation characteristics;

And step six, classifying and sorting all issuing sequences, eliminating repeated and unreasonable issuing moment vectors, combining issuing waveforms of the same muscle movement unit, calculating a waveform mean value, and optimizing a result.

2. The method for extracting a muscle movement unit based on a neural network as claimed in claim 1, wherein in step four, the classification template is dynamically updated by using the change of the issued waveform in the dynamic contraction process, and the specific updating formula is as follows:

z_i+1＝ηz_i+(1-η)s_i

Wherein z is_i+1Is the updated template, z_iis the template before update, η is the update coefficient, s_iIs a factor that changes the dynamic contraction of muscles, i.e.p_iIs the waveform of the dynamic input variation and N is the waveform length.

Technical Field

The invention relates to a muscle movement unit extraction method based on a neural network.

Background

Surface electromyogram (surface EMG) signals are detected from the body surface of a human body by using surface electrodes, and compared with Needle electrode electromyogram (NEMG), the surface electromyogram (surface EMG) signals have the characteristics of no wound and easy acceptance, so the surface electromyogram (surface EMG) signals have wide application prospects, and are particularly used for detection and analysis of array type sEMG signals. Clinically, the functional state of the neuromuscular can be known more comprehensively through the array type sEMG. At present, the sEMG signal analysis and processing mainly aims at the surface electromyogram signals of static muscle contraction, the research on dynamic muscle contraction is very few, and the invention provides a muscle Motor Unit (MU) extraction method based on a neural network aiming at the dynamic muscle contraction.

Disclosure of Invention

In view of the above problems, an object of the present invention is to provide a neural network-based muscle motor unit extraction method, including the steps of:

collecting array type surface electromyographic signals when muscles contract dynamically, filtering the signals and weakening interference;

Step two, extracting the muscle movement unit on each channel to send out waveforms;

Inputting the waveform emitted by the muscle movement unit of each channel into an ART2 neural network, and setting a nonlinear function in a neuron to suppress noise;

Step four, in the ART2 recognition and classification process, dynamically updating the classification template, comparing the distribution waveform similarity and outputting the classification result;

Step five, carrying out differential operation on the classification results output by each channel, and confirming the muscle movement unit according to the propagation characteristics;

and step six, classifying and sorting all issuing sequences, eliminating repeated and unreasonable issuing moment vectors, combining issuing waveforms of the same muscle movement unit, calculating a waveform mean value, and optimizing a result.

The optimization measures comprise:

In the fourth step, the classification template is dynamically updated by using the change of the issued waveform in the dynamic contraction process, and the specific updating formula is as follows:

z_i+1＝ηz_i+(1-η)s_i

Wherein z is_i+1Is the updated template, z_iIs the template before update, η is the update coefficient, s_iIs a factor that changes the dynamic contraction of muscles, i.e.p_iIs the waveform of the dynamic input variation and N is the waveform length.

Compared with the prior art, the muscle movement unit extraction method based on the neural network provided by the invention has the advantages that the neural network training is not needed in the extraction process, the actual waveform signal is directly input for identification, the identification process is simplified, and the operation time is saved. When the new input waveform is different, if the waveform change is within the range of the dynamic contraction change, the existing template is assigned and the template is updated, otherwise, the new template is generated. When the muscle movement unit is confirmed, the waveform propagation characteristic is considered, and the accuracy of extracting the muscle movement unit is improved.

drawings

FIG. 1 is a flow chart of the present invention.

Fig. 2 is a schematic diagram of the position of a muscle movement unit according to an embodiment of the invention.

Detailed Description

the present invention is further described in detail below with reference to the accompanying drawings, and can be easily implemented by those skilled in the art from the disclosure of the present specification.

Fig. 1 shows a flow chart of the present invention.

the invention provides a muscle movement unit extraction method based on a neural network, which is characterized by comprising the following steps of:

collecting array type surface electromyographic signals when muscles contract dynamically, filtering the signals and weakening interference. Because the sEMG signal contains various interference signals, the preprocessing firstly needs to adopt a band-pass filter, reserves signals of a frequency range of 10 Hz-500 Hz, and then adopts a notch filter to filter 50Hz power frequency interference.

And step two, extracting the muscle movement unit release waveform on each channel. And during extraction, setting a peak threshold value which is higher than the threshold value, extracting waveforms with the length of 40ms before and after the peak time, and otherwise, not extracting.

and step three, inputting the muscle movement unit sending waveform of each channel into an ART2 neural network, and setting a nonlinear function in the neuron to suppress noise. Firstly, initializing ART2 neural network, giving initial value to parameter, then inputting the waveform of muscle movement unit to be classified, finally calculating ART2 comparing layer and identifying layer vector, finding out winning node. Setting a nonlinear function in the neuron to suppress noise, wherein the nonlinear function is as follows:

where θ ∈ (0,1) is a constant and x is the input dispensing waveform.

And step four, in the ART2 identification and classification process, dynamically updating the classification template, comparing the distribution waveform similarity and outputting the classification result. The ART2 neural network returns the top-down vector from the winning node of the recognition layer, and calculates the degree of similarity. The classification template is dynamically updated by using the change of the issued waveform in the dynamic contraction process, and the specific updating formula is as follows:

z_i+1＝ηz_i+(1-η)s_i

wherein z is_i+1Is the updated template, z_iis the template before update, η is the update coefficient, s_iIs a factor that changes the dynamic contraction of muscles, i.e.p_iIs the waveform of the dynamic input variation and N is the waveform length.

Calculating the similarity degree of the input muscle movement unit distribution waveform and the template according to the updated classification template, receiving or not receiving the input distribution waveform as a winning node, and adjusting the top vector and the bottom vector of the ART2 neural network.

and fifthly, carrying out differential operation on the classification results output by the channels, and confirming the muscle movement unit according to the propagation characteristics. And (3) carrying out difference operation on the waveforms on the electrodes according to the corresponding relation of the time sequence, as shown in figure 2, determining the muscle movement unit according to the characteristics that the signals on the graph and the muscle movement unit distribution waveforms are transmitted to two sides, the waveform of the terminal plate area is minimum, and the position of the left mark in figure 2 is the terminal plate area of the muscle movement unit.

and step six, classifying and sorting all issuing sequences, eliminating repeated and unreasonable issuing moment vectors, combining issuing waveforms of the same muscle movement unit, and calculating a waveform mean value.

In conclusion, the ART2 neural network is adopted to classify the waveforms of the muscle movement units, the neural network training is not needed in the classification process, and the actual waveform signals are directly input for identification, so that the identification process is simplified, and the running time is saved. In the classification process, the classification template is dynamically updated in order to deal with the change of the motion unit when the muscle is dynamically contracted. When the muscle movement unit is confirmed, the waveform propagation characteristics are considered, and the accuracy of extracting the muscle movement unit is improved. The invention is simple to realize and meets the requirement of practical application.

the foregoing embodiments are merely illustrative of the principles and utilities of the present invention and are not intended to limit the invention. Any person skilled in the art can modify or change the above-mentioned embodiments without departing from the spirit and scope of the present invention. Accordingly, it is intended that all equivalent modifications or changes which can be made by those skilled in the art without departing from the spirit and technical spirit of the present invention be covered by the claims of the present invention.