阅读说明：本技术 一种多通道肌电特征图像的构建方法 (Construction method of multi-channel electromyographic feature image ) 是由 李公法 刘鑫 江都 陶波 蒋国璋 孔建益 孙瑛 童锡良 于 2021-09-13 设计创作，主要内容包括：本发明是一种多通道肌电特征图像数据集的构建方法。首先,结合阈值比较法和巴特沃斯滤波器,完成原始肌电信号的预处理。随后,从常用的时域和频域特征中进行2次筛选,选择4种互不冗余的肌电特征进行特征提取。其次,基于一维信号与二维图像的映射关系生成肌电图像。最后,按照图像的拼接方式,完成多通道肌电特征图像的构建。肌电图像数据集通过深度学习网络的训练,可以有效提高手势识别率。多通道肌电特征图像具有更加丰富的特征信息,多种特征可以完成信息之间的互补,最后的识别率会比单通道肌电特征图像高出8％-9％。(The invention relates to a construction method of a multi-channel electromyographic feature image data set. Firstly, a threshold comparison method and a Butterworth filter are combined to finish the preprocessing of the original electromyographic signals. Then, 2 times of screening is carried out from the common time domain and frequency domain characteristics, and 4 non-redundant electromyographic characteristics are selected for characteristic extraction. And secondly, generating an electromyographic image based on the mapping relation between the one-dimensional signal and the two-dimensional image. And finally, completing the construction of the multi-channel electromyogram characteristic image according to the splicing mode of the image. The myoelectric image data set can effectively improve the gesture recognition rate through the training of the deep learning network. The multi-channel electromyographic characteristic image has richer characteristic information, multiple characteristics can complete complementation between information, and the final recognition rate is 8% -9% higher than that of a single-channel electromyographic characteristic image.)

1. A construction method of a multi-channel electromyographic feature image is characterized by comprising the following steps:

step 1: acquiring electromyographic signals sEMG of the surface of a forearm of a human body, and finishing the pretreatment work of extracting an active section and filtering and denoising;

step 2: selecting proper time domain characteristics and frequency domain characteristics;

and step 3: generating single-channel electromyographic feature Image based on mapping relation between one-dimensional signal and two-dimensional Image_{Single channel}；

And 4, step 4: comparing the images with the preliminarily screened features, and screening out 4 kinds of feature images by utilizing similarity analysis;

and 5: the multi-channel electromyographic feature Image is completed by combining an Image channel splicing method_{Multi-channel}Wherein the 4 screened characteristic images are 4 channels thereof;

step 6: and (5) repeating the steps 3 to 5, completing the conversion of the signals into the pictures and constructing a multi-channel myoelectricity characteristic image data set.

2. The method for constructing the multi-channel electromyographic feature image according to claim 1, wherein the step 1 comprises the following steps of extracting an active segment S (n) by using a threshold comparison method:

wherein C is the collection channel of sEMG, N is the number of sampling points, sEMG_c(n) is the value of the nth sample point of the c channel,the mean value of sEMG when the c-channel muscle is relaxed, and TH is a set threshold;

low pass filtering is performed by a first order butterworth filter of 1HZ,

wherein the content of the first and second substances,in order to be the amplitude of the vibration,is the frequency, epsilon is the scaling factor, n is the order of the filter,in order to cut-off the frequency of the frequency,the passband edge frequency.

3. The method as claimed in claim 1, wherein the step 3 is implemented by using a single-channel electromyographic feature Image_{Single channel}Comprises the following steps:

Image_{single channel}∈S^W×H×1

W represents the width of a sliding window, namely the width of an image, H represents the number of channels of an original electromyographic signal, and also represents the height of the image or the sliding window, the sliding window is used for completing the mapping of a one-dimensional signal to a gray image in a [0, 1] interval, and the mapping relation is as follows:

A_[0,1]＝F(A_(i,j))

F(A_(i,j))＝(A_(i,j)-A_min)/(A_max-A_min)

a is a sEMG matrix formed after sEMG passes through a sliding window, A_[0,1]To turn toThe converted electromyogram, F () is a Fourier function, A_(i,j)Is the value of the ith row and the jth column in the sEMG matrix, wherein 0<i<H，0<j<W，A_maxIs the largest value in the sEMG matrix, A_minIs the smallest value in the sEMG matrix.

4. The method as claimed in claim 1, wherein the multi-channel electromyographic feature Image in step 5 is constructed by_{Multi-channel}Comprises the following steps:

Image_{multi-channel}∈S^W×H×4

Each channel of the multi-channel electromyography characteristic image is composed of 4 single-channel electromyography characteristic images, the size of each single-channel electromyography characteristic image is W multiplied by H, C is the number of channels of the image and is also the number of the characteristics, the size of the multi-channel electromyography characteristic image formed by combining the single-channel electromyography characteristic images is W multiplied by H multiplied by C, and C is 4.

Technical Field

The invention relates to the field of signal processing and man-machine interaction, in particular to a bidirectional recursive myoelectric feature selection method based on myoelectric signals.

Background

With the development of science and technology, human-computer interaction research and application are concerned by more and more scholars. Among them, gesture recognition is one of the main research topics in the field of human-computer interaction as a main branch of human-computer interaction. Gestures in human-computer interaction have the advantages of convenience, strong interactivity, rich expression content and the like for users, and are always the first choice in the field of human-computer interaction. The electromyographic signal is a bioelectric signal generated in the process of human muscle activity and is a comprehensive result of conduction and superposition of different movement unit action potentials in muscle fibers and cell tissues. The myoelectric signal is mainly used for gesture recognition by extracting myoelectric features, but the myoelectric gesture classification recognition rate based on the traditional method is always in a stagnation state.

In view of the above, a deep learning method is required to be combined for myoelectric gesture recognition, a traditional method is combined with deep learning, myoelectric features of different representations are extracted by using the traditional method to generate myoelectric images, more abstract information contained in the feature map is understood through the deep learning, and finally, information fusion is completed to obtain better classification performance.

Disclosure of Invention

The invention provides a method for constructing a multi-channel electromyographic feature image, which can complete the conversion of a single-channel electromyographic feature image from various electromyographic features through electromyographic feature extraction and the mapping mathematical relationship from a one-dimensional signal to a two-dimensional image, and then complete the construction of the multi-channel electromyographic feature image based on a splicing method of image channels.

In order to solve the technical problems, according to the invention, the following technical scheme is adopted as a method for constructing a multi-channel myoelectric characteristic image, and the method specifically comprises the following steps:

step 1: acquiring electromyographic signals sEMG of the surface of a forearm of a human body, and finishing the pretreatment work of extracting an active section and filtering and denoising;

step 2: selecting proper time domain characteristics and frequency domain characteristics;

and step 3: generating single-channel electromyographic feature Image based on mapping relation between one-dimensional signal and two-dimensional Image_{Single channel}；

And 4, step 4: comparing the images with the preliminarily screened features, and screening out 4 kinds of feature images by utilizing similarity analysis;

and 5: the multi-channel electromyographic feature Image is completed by combining an Image channel splicing method_{Multi-channel}Wherein the 4 screened characteristic images are 4 channels thereof;

step 6: and (5) repeating the steps 3 to 5, completing the conversion of the signals into the pictures and constructing a multi-channel myoelectricity characteristic image data set.

Further, in step 1, the threshold comparison method is used to extract an active segment s (n):

wherein C is the collection channel of sEMG, N is the number of sampling points, sEMG_c(n) is the value of the nth sample point of the c channel,mean sEMG at c-channel muscle relaxation, TH is the set threshold.

Low pass filtering is performed by a first order butterworth filter of 1HZ, wherein,in order to be the amplitude of the vibration,is the frequency, epsilon is the scaling factor, n is the order of the filter,in order to cut-off the frequency of the frequency,the passband edge frequency.

Further, in step 3, the single-channel electromyography characteristic Image_{Single channel}Comprises the following steps:

Image_{single channel}∈S^W×H×1

W represents the width of the sliding window, which is also the width of the image, and H represents the number of channels of the raw electromyogram signal, which also represents the image or the height of the sliding window. And (3) completing the mapping of the one-dimensional signal to the gray level image in the [0, 1] interval by using a sliding window, wherein the mapping relation is as follows:

A_[0,1]＝F(A_(i,j))

F(A_(i,j))＝(A_(i,j)-A_min)/(A_max-A_min)

a is a sEMG matrix formed after sEMG passes through a sliding window, A_[0,1]For the converted electromyogram F () is a Fourier function, A_(i,j)Is the value of the ith row and the jth column in the sEMG matrix, wherein 0<i<H，0<j<W，A_maxIs the largest value in the sEMG matrix, A_minIs the smallest value in the sEMG matrix.

Further, in step 5, the multi-channel electromyographic feature Image_{Multi-channel}Comprises the following steps:

Image_{multi-channel}∈S^W×H×4

Each channel of the multi-channel electromyography characteristic image is composed of 4 single-channel electromyography characteristic images, the size of each single-channel electromyography characteristic image is W multiplied by H, C is the number of channels of the image and is also the number of the characteristics, the size of the multi-channel electromyography characteristic image formed by combining the single-channel electromyography characteristic images is W multiplied by H multiplied by C, and C is 4.

Compared with the prior art, the invention has the beneficial effects that:

the method for constructing the multi-channel electromyographic feature image provided by the invention is based on the mapping relation between signals and images and the image channel splicing principle, converts one-dimensional electromyographic signals into two-dimensional electromyographic images, and is beneficial to constructing an electromyographic feature image data set. The myoelectric image data set can effectively improve the gesture recognition rate through the training of the deep learning network. The multi-channel electromyographic characteristic image has richer characteristic information, multiple characteristics can complete complementation between information, and the final recognition rate is 8% -9% higher than that of a single-channel electromyographic characteristic image.

Drawings

FIG. 1: is a flow chart of the method of the present invention;

FIG. 2: a gesture graph for collecting electromyographic signals;

FIG. 3: constructing a schematic diagram of a multi-channel electromyographic feature image;

FIG. 4: is an electromyographic signal acquisition device.

Detailed Description

The technical solutions in the embodiments of the present invention will be 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 of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.

The specific implementation mode of the invention adopts the following technical scheme to be a method for constructing a multi-channel electromyographic characteristic image, which specifically comprises the following steps:

step 1: the myoelectric signal of the forearm surface acting as shown in fig. 2 is acquired by the myoelectric signal acquisition device shown in fig. 4, the 9 gestures comprise palm closing (SH) and palm opening (SK), the wrist movement comprises wrist flexion (NQ) and external flexion (WQ), the finger movement comprises the acting force of the thumb on the index finger (MS), the middle finger (MZ), the ring finger (MW) and the little finger (MX), and besides, the test gesture action also comprises a rest action (RE). And after electromyographic signal acquisition is finished, preprocessing work such as signal active section extraction, filtering and noise reduction is carried out. The active segment refers to a muscle contraction signal in the action execution process, and the purpose of active segment extraction is to remove the inactive segment which does not contain muscle contraction;

in the step 1, an active segment S (n) is extracted by using a threshold comparison method:

wherein C is the collection channel of sEMG, N is the number of sampling points, sEMG_c(n) is the value of the nth sample point of the c channel,mean sEMG at c-channel muscle relaxation, TH is the set threshold.

Low pass filtering is performed by a first order butterworth filter of 1HZ, wherein,in order to be the amplitude of the vibration,is the frequency, epsilon is the scaling factor, n is the order of the filter,in order to cut-off the frequency of the frequency,the passband edge frequency.

Step 2: 4 myoelectric characteristics are selected, which are respectively as follows: root Mean Square (RMS), Waveform Length (WL), sample entropy (SampEn), and Median Amplitude Spectrum (MAS);

and step 3: generating a single-channel electromyographic feature image based on the mapping relation between the one-dimensional signal and the two-dimensional image;

step 3, the single-channel electromyographic characteristic Image_{Single channel}Comprises the following steps:

Image_{single channel}∈S^W×H×1

W represents the width of the sliding window, which is also the width of the image, and H represents the number of channels of the raw electromyogram signal, which also represents the image or the height of the sliding window. And (3) completing the mapping of the one-dimensional signal to the gray level image in the [0, 1] interval by using a sliding window, wherein the mapping relation is as follows:

A_[0,1]＝F(A_(i,j))

F(A_(i,j))＝(A_(i,j)-A_min)/(A_max-A_min)

a is a sEMG matrix formed after sEMG passes through a sliding window, A_[0,1]For the converted electromyogram F () is a Fourier function, A_(i,j)Is the value of the ith row and the jth column in the sEMG matrix, wherein 0<i<H，0<j<W，A_maxIs the largest value in the sEMG matrix, A_minIs the smallest value in the sEMG matrix.

And 4, step 4: comparing the images with the preliminarily screened features, and screening out 4 kinds of feature images by utilizing similarity analysis;

and 5: and finishing the multi-channel electromyography characteristic images by combining an image channel splicing method, wherein the 4 screened characteristic images are 4 channels thereof.

Step 5, the multi-channel electromyographic characteristic Image_{Multi-channel}Comprises the following steps:

Image_{multi-channel}∈S^W×H×4

Each channel of the multi-channel electromyogram characteristic image is composed of 4 single-channel electromyogram characteristic images, the size of each single-channel electromyogram characteristic image is W multiplied by H, C is the number of channels of the image and is also the characteristic quantity, the size of the multi-channel electromyogram characteristic image formed by combining the single-channel electromyogram characteristic images is W multiplied by H multiplied by C, C is 4, and the generation principle is shown in fig. 3.

Although the present invention has been described in detail with reference to the foregoing embodiments, it will be apparent to those skilled in the art that various changes, modifications, equivalents, and improvements may be made without departing from the spirit and scope of the invention.