A laryngeal muscle electromyographic signal acquisition and processing method based on surface electromyography

Abstract

The present application belongs to the technical field of muscle electrical signal acquisition and processing, and relates to a laryngeal muscle electrical signal acquisition and processing method based on surface electromyography. The method is a non-invasive method that can solve the disadvantage of LEMG invasiveness by collecting the electrical activity signals of laryngeal muscles through sEMG technology and a multi-channel sensor placement method, and provides a basis for decomposing target muscle electrical activity in the signal processing process through the sensor placement method, thereby solving the problem that the existing sEMG acquisition method cannot obtain laryngeal target muscle electrical activity signals. Then, rectification, filtering, wavelet transform, Fast ICA decomposition, inverse Fast ICA and inverse wavelet are used to realize the denoising of all channel surface electromyography signals, and the inhibition effect on electrocardiogram interference, channel noise, system noise and other noises is remarkable.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of muscle electrical signal acquisition and processing, and relates to a laryngeal muscle electrical signal acquisition and processing method based on surface electromyography. BACKGROUND

[0002] The larynx is an important organ composed of a complex tubular structure formed by cartilages, joints, muscles and ligaments. The driving force of laryngeal movement is the laryngeal muscle. Under the correct innervation, the normal coordinated movement of the laryngeal muscle ensures the correct execution of the laryngeal function. Abnormal laryngeal nerve function or abnormal activity of the laryngeal muscle has a significant impact on the laryngeal function, which can seriously affect the quality of life of patients. The measurement of laryngeal muscle electrical signal can directly reflect the movement state of laryngeal muscle and indirectly reflect the control state of laryngeal nerve, which has important basic research value and clinical application value for exploring the causes and dynamic changes of laryngeal diseases.

[0003] The existing laryngeal muscle electrical signal acquisition and processing includes the following ways:

[0004] Muscle electrical signal acquisition:

[0005] At present, the most commonly used method for detecting laryngeal muscle activity in clinic is laryngeal electromyography (LEMG), which has little risk to the subject and requires the operator to be familiar with the precise anatomical structure of the laryngeal muscle and have the necessary electrophysiological knowledge. During detection, small electrodes such as needle-shaped and hook-shaped electrodes are inserted into the laryngeal muscle, which can better stabilize the contact with the measured muscle and the collected muscle electrical signal is less disturbed by skin, fat and other tissues, which can be used to evaluate various laryngeal diseases and laryngeal muscle and laryngeal nerve function.

[0006] Some studies also use surface electromyography (sEMG) technology to collect laryngeal muscle activity electrical signals to achieve non-invasive measurement of laryngeal muscle electrical activity.

[0007] Muscle electrical signal processing:

[0008] The common processing methods of muscle electrical signal include: using ordinary filters (such as Chebyshev filter, notch filter, Butterworth filter, etc.) to filter out stop-band noise; using moving median filter and moving average filter for smoothing processing of the signal; using adaptive filter to automatically iteratively adjust its filter parameters to achieve optimal filtering of noise; using wavelet transform to locate and remove noise; using principal component analysis (PCA) to remove noise containing less effective information; using independent component analysis (ICA) to separate noise components.

[0009] However, the existing laryngeal muscle electrical signal acquisition and processing methods all have different kinds of defects, which are as follows:

[0010] Signal acquisition:

[0011] Laryngeal electromyography (LEMG) is an invasive technique, which is invasive, and the electrode needle needs to be penetrated from the larynx or orally entered and continuously adjusted in angle to find the appropriate electrode position, which requires higher technical and experience of the operator and brings greater pain and poor comfort to the subject.

[0012] Although the sEMG technology is non-invasive, the laryngeal muscles are small, distributed in a concentrated manner and the surface sensor is relatively large in size, the sEMG signal collected is inevitably affected by the surrounding interference muscles of the target muscle, and the existing sEMG collection scheme cannot obtain the signal reflecting the electrical activity of the laryngeal target muscle.

[0013] In terms of signal processing:

[0014] The use of ordinary filters inevitably has a transition band, and part of the noise is still not effectively filtered out; the average filter and the median filter have poor filtering ability for low-frequency noise; the use of adaptive filtering requires the assumption that the noise and the reference signal are in a linear relationship; the wavelet transform denoising needs to locate the noise signal, and without prior noise information, effective denoising cannot be achieved; the number of potential principal components of PCA cannot be well estimated, so the data information cannot be well preserved; the number of independent component signals obtained by ICA decomposition is the same as the number of signal channels used for ICA decomposition, that is, if there are only two-channel signals, only two independent components that are independent of each other can be decomposed, and therefore the number of channels is very important for whether the noise can be decomposed.

[0015] Therefore, a more optimal laryngeal muscle electrical signal collection and processing method is needed to solve the defects of the laryngeal muscle electrical signal collection and processing method in the prior art. SUMMARY

[0016] The technical solution adopted by the present application to solve the technical problems is: a laryngeal muscle electrical signal collection and processing method based on surface electromyography, comprising the following steps:

[0017] Step 1: Collecting sEMG signals of laryngeal target muscles by using different multi-channel differential sensor placement methods, that is, collecting electrical activities of different target muscles and their surrounding interference muscles through different multi-channel (channel number ≥ 2, that is, using two or more sensors) differential sensor special placement positions and directions (one multi-channel differential sensor placement method collects sEMG signals of one laryngeal target muscle). Taking the laryngeal target muscles cricothyroid muscle (CT muscle) and lateral cricoarytenoid muscle (LCA muscle) as examples, through sensors in different positions and directions, sEMG signals of the CT muscle can be collected non-invasively, sEMG signals of the LCA muscle can be collected non-invasively, and sEMG signals of the CT muscle and the LCA muscle can be collected non-invasively at the same time.

[0018] Specifically, the sensor placement mode for collecting the sEMG signal of the CT muscle is mode 1, and the sensor placement mode for collecting the sEMG signal of the LCA muscle is mode 2. Mode 1 requires that the main sensor of at least one channel is placed on the CT muscle and parallel to the axial direction of the CT muscle, for collecting the sEMG signal of the CT muscle; and requires that the auxiliary sensor of at least one channel is placed close to the upper edge or lower edge of the main sensor and parallel to the main sensor, for collecting the sEMG signal of the interfering muscle around the CT muscle. Mode 2 requires that the main sensor of at least one channel is placed on the LCA muscle and parallel to the axial direction of the LCA muscle, for collecting the sEMG signal of the LCA muscle; and requires that the auxiliary sensor of at least one channel is placed close to the upper edge or lower edge of the main sensor and parallel to the main sensor, for collecting the sEMG signal of the interfering muscle around the LCA muscle. The sensor placement mode provided by the present application can realize non-invasive measurement of the activity of the laryngeal muscle, and at the same time, the sensor position and direction provide a basis for decomposing the target muscle electrical signal in the subsequent signal processing process, solving the problem that the existing sEMG acquisition method cannot obtain the signal reflecting the electrical activity of the target muscle in the larynx;

[0019] Step two: rectify the sEMG signal collected in step one into a zero-mean sEMG signal:

[0020] Step three: pass the zero-mean sEMG signal obtained in step two through a band-pass filter to limit the sEMG signal spectrum within the sEMG signal frequency band and filter out low-frequency and high-frequency noise;

[0021] Step four: perform wavelet transform on the single-channel signal filtered by the band-pass filter in step three to decompose the single-channel signal into multiple groups of signals with different frequency components;

[0022] Step five: decompose the multiple-component signals obtained by wavelet transform in step four using the Fast ICA algorithm to obtain the same number of independent component (IC) signals, and then selectively remove noise components, including components with significant electrocardiogram signals and channel noise;

[0023] Step six: perform inverse Fast ICA and inverse wavelet transform on the IC signals remaining after removing the noise components in step five to restore the multi-channel IC signals to single-channel signals, which are the denoised single-channel sEMG signals;

[0024] Step seven: repeat steps two to six to the sEMG signals of different channels until all the denoised sEMG signals of different channels are obtained; by repeating steps two to six to rectify, filter, wavelet transform, Fast ICA decomposition, inverse Fast ICA and inverse wavelet transform, the denoising of all surface electromyography signals is realized, and the suppression effect of ECG interference, channel noise, system noise and other noises is remarkable;

[0025] Step eight: use Fast ICA algorithm to decompose the multi-channel (≥2) sEMG signals (sEMG signals collected by the main sensor of at least one channel and the auxiliary sensor of at least one channel) obtained in step seven, separate the sEMG signals of the target muscle and the surrounding interference muscle in the larynx, and obtain the sEMG signals of the target muscle and the surrounding interference target muscle, respectively, denoted as IC1, IC2, …, ICn. n , wherein n is the number of channels of the multi-channel sEMG signal;

[0026] Step nine: calculate the correlation coefficient between IC i (i = 1, …, n) obtained in step eight and the sEMG signals of each channel; find the IC i that meets the following correlation principle: it is believed that the sEMG signal collected by the main sensor contains more laryngeal target muscle electrical activity signal, so the IC i that meets the correlation coefficient greater than 0.8 with the sEMG signal collected by the main sensor is found, and the correlation coefficient less than 0.5 with the sEMG signal collected by the auxiliary sensor is met, and this IC i is denoted as IC_target, that is, the sEMG signal of the laryngeal target muscle extracted by Fast ICA decomposition. If no IC i that meets the condition is found, remove this set of multi-channel sEMG signals and do not make further analysis;

[0027] Step ten: extract at least one time-frequency domain sEMG signal feature parameter in a series of sliding rectangular analysis windows for the laryngeal target muscle sEMG signal IC_target extracted in step nine.

[0028] Preferably, in step one, when the target muscle is the cricothyroid muscle, at least one main sensor is required to be parallel to the axial direction of the cricothyroid muscle and located on the cricothyroid muscle for acquiring the sEMG signal of the cricothyroid muscle. At least one secondary sensor is required to be parallel to the main sensor and placed close to the upper or lower edge of the main sensor for acquiring the sEMG signal of the interfering muscles around the cricothyroid muscle. When the target muscle is the cricoarytenoid muscle, at least one main sensor is required to be parallel to the direction of the cricoarytenoid muscle and located on the cricoarytenoid muscle for acquiring the sEMG signal of the cricoarytenoid muscle. At least one secondary sensor is required to be parallel to the main sensor and placed close to the upper or lower edge of the main sensor for acquiring the sEMG signal of the interfering muscles around the cricoarytenoid muscle. By performing Fast ICA on the denoised signal, an independent component equal to the number of channels is obtained. Then, based on the correlation principle, the independent component that better reflects the electrical activity of the target muscle is selected to remove the influence of the surrounding interfering muscles to the greatest extent, and the time-frequency domain features of at least one surface electromyography of the independent component are extracted.

[0029] Preferably, in step one, the target muscles further include: the thyroarytenoid muscle, the sternothyroid muscle, the sternohyoid muscle, and the thyrohyoid muscle.

[0030] Preferably, in step nine, the IC obtained in step eight... i The search for the cricothyroid muscle or cricoarytenoid muscle, the target muscles of the larynx, in the range i = 1, ..., n, is conducted by calculating the IC... i The correlation coefficient between the IC and the sEMG signals of each channel; find ICs that meet the following correlation criteria. i It is believed that the sEMG signal acquired by the main sensor contains more electrical activity signals of the target muscles in the throat. Therefore, we are looking for an IC that has a correlation coefficient greater than 0.8 with the sEMG signal acquired by the main sensor. i And satisfying the condition that the correlation coefficient between the IC and the sEMG signal acquired by the secondary sensor is less than 0.5, this IC is denoted as IC. i IC_target is the sEMG signal of the target laryngeal muscle extracted by Fast ICA decomposition; if no matching IC is found... i If so, this set of multi-channel sEMG signals is removed and no further analysis is performed.

[0031] Preferably, in step three, the upper passband frequency of the bandpass filter is selected in the range of 350-500Hz, the lower passband frequency is selected in the range of 0-50Hz, the upper stopband frequency is 500Hz, and the lower stopband frequency is selected in the range of 0-20Hz.

[0032] Preferably, in step four, the wavelet function can be selected as the Symlet wavelet or the Daubechies wavelet, and the decomposition scale can be selected in the range of 6-10.

[0033] Preferably, in step ten, the time-domain sEMG parameters are calculated according to the following formula:

[0034]

[0035]

[0036]

[0037]

[0038]

[0039] wherein s is the sEMG signal, and N is the data length.

[0040] The frequency-domain sEMG parameters are calculated according to the following formula:

[0041]

[0042]

[0043]

[0044]

[0045]

[0046]

[0047]

[0048]

[0049] wherein P is the power spectral density, f is the corresponding frequency, and N is the data length.

[0050] Preferably, in step ten, the window length of the sliding rectangular analysis window is selected in the range of 20-500 ms, and the step length can be 30%-60% of the window length.

[0051] Preferably, in step ten, the window length of the sliding rectangular analysis window is 50 ms, and the step length is 15 ms.

[0052] Preferably, in step ten, the time-domain sEMG signal feature parameters include the root mean square value RMS, the absolute mean value MAV, the integral value INTE, the variance VAR, and the zero-crossing rate ZC; and the frequency-domain sEMG signal feature parameters include the median frequency MDF, the average power frequency MPF, the frequency ratio FR, the total power TTP, the 1-3 order spectral moments SM1-SM3, and the central frequency variance VCF.

[0053] The beneficial effects of the present application are:

[0054] 1. The laryngeal muscle sEMG signal acquisition method provided by the present application acquires the electrical activity signal of the laryngeal muscle through the sEMG technology and the multi-channel sensor placement method, is a non-invasive method, can solve the disadvantage of LEMG invasiveness, and provides a basis for decomposing the laryngeal target muscle electrical activity in the signal processing process through the sensor placement method, and solves the problem that the existing laryngeal muscle sEMG acquisition method cannot obtain the laryngeal target muscle electrical activity signal.

[0055] 2. The signal processing method provided by the present application, (1) first rectifies to obtain a zero-mean signal, and then removes high-frequency and low-frequency noise through a band-pass filter, overcoming the limitations of average and median filters in filtering low-frequency noise; (2) through wavelet transform, Fast ICA decomposition and their inverse processes in sequence, noise (residual electrocardio interference, residual high-frequency noise and residual low-frequency disturbance) not filtered by the band-pass filter is filtered out; (3) when removing electrocardio interference through wavelet transform, Fast ICA decomposition and their inverse processes, reference electrocardio does not need to be collected, the linear relationship between electrocardio interference and reference electrocardio does not need to be assumed, and electrocardio does not need to be positioned, overcoming the limitations of wavelet transform and adaptive filtering; (4) when removing noise through wavelet transform, Fast ICA decomposition and their inverse processes, the single-channel signal is decomposed into a multi-channel signal through wavelet transform, the multi-channel signal is then decomposed through Fast ICA, noise components are selectively removed, and the multi-component signal is restored to a single-channel signal through the inverse process, as many data information as possible is retained, and the problems of PCA data information loss and high requirement of ICA on the number of channels are solved; (5) through Fast ICA decomposition of the denoised signal and the correlation principle adopted by the present application, the signal that can more reflect the target muscle electrical activity can be screened out, and the influence of the surrounding interference muscle is removed to the greatest extent; (6) because the influence of noise and the surrounding interference signal is weakened to the greatest extent and more effectively, the dynamic changes of the more accurate sEMG time-frequency domain feature parameters can be extracted.

[0056] 3. Through the acquisition and signal processing method of the present application, the laryngeal muscle electrical activity signal can be effectively extracted, and the problems that LEMG is invasive and the existing sEMG acquisition method cannot separate the laryngeal target muscle and the surrounding interference muscle electrical activity are overcome. BRIEF DESCRIPTION OF DRAWINGS

[0057] Figure 1 It is a flowchart of a laryngeal muscle electrical signal acquisition and processing method based on surface electromyography; DETAILED DESCRIPTION

[0058] The related technologies in the present application will be described clearly and completely in connection with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all the other embodiments obtained by those skilled in the art without creative work belong to the protection scope of the present application.

[0059] Reference Figure 1 A laryngeal muscle electrical signal acquisition and processing method based on surface electromyography, comprising:

[0060] Step one: different multi-channel differential sensor placement methods are used to collect sEMG signals of different laryngeal muscles. That is, different multi-channel (channel number ≥ 2) differential sensor special position placement positions and directions are used to collect the electrical activity of different laryngeal target muscles and the electrical activity signals of the surrounding interference muscles (a multi-channel differential sensor placement method corresponds to the sEMG signal collection of a laryngeal target muscle). A certain muscle can be collected alone, or two or more muscles can be collected together. By collecting sEMG signals of laryngeal muscles through differential sensors in different positions and directions, non-invasive measurement of laryngeal muscle activity can be realized, and the special placement of the sensor provides a basis for extracting target muscle electrical signals in the subsequent signal processing process. Taking the collection of sEMG signals of important laryngeal muscles CT muscle and LCA muscle as an example, when collecting sEMG signals of CT muscle: at least one channel of the main sensor is required to be placed on the CT muscle and parallel to the axial direction of the CT muscle, which is used to collect the sEMG signals of the CT muscle; at least one channel of the auxiliary sensor is required to be placed parallel to the main sensor and close to the upper edge or lower edge of the main sensor, which is used to collect the sEMG signals of the surrounding interference muscles (such as sternohyoid muscle, sternothyroid muscle, thyrohyoid muscle, and levator scapulae muscle); when collecting sEMG signals of LCA muscle: at least one channel of the main sensor is required to be placed on the LCA muscle and parallel to the axial direction of the LCA muscle, which is used to collect the sEMG signals of the LCA muscle; at least one channel of the auxiliary sensor is required to be placed parallel to the main sensor and close to the upper edge or lower edge of the main sensor, which is used to collect the sEMG signals of the surrounding interference muscles (such as sternohyoid muscle, sternothyroid muscle, thyrohyoid muscle, and levator scapulae muscle).

[0061] Step two: the sEMG signals collected in step one are rectified to zero mean sEMG signals by subtracting the mean value:

[0062]

[0063] Wherein, is the rectified zero mean signal of sEMG signal, s is sEMG signal, and N is data length.

[0064] Step three: the zero-mean sEMG signal obtained in step two is filtered through a butterworth band-pass filter to filter out low-frequency and high-frequency noise and limit the signal frequency band within the sEMG signal frequency band range;

[0065] After step three, the electrocardiogram interference signal can still be observed in the sEMG signal, therefore, steps four to seven are used to remove the electrocardiogram signal;

[0066] Step four: the single-channel signal obtained in step three is subjected to wavelet transform to decompose the single-channel signal into a plurality of signals of different frequency components;

[0067] Step five: the plurality of component signals obtained in step four are decomposed using the Fast ICA algorithm to obtain the same number of independent component signals, and then noise components are selectively removed, the noise components including but not limited to electrocardiogram interference signals and channel noise;

[0068] Step six: the independent component signals remaining after the noise components are removed in step five are subjected to inverse Fast ICA and inverse wavelet transform in sequence to restore the plurality of component signals to a single-channel signal, which is a single-channel sEMG signal after denoising;

[0069] Step seven: steps two to six are repeated for the sEMG signals of different channels until the sEMG signals of all channels after denoising are obtained; the denoising of the sEMG signals of all channels is achieved by repeating steps two to six for signal rectification filtering, wavelet transform, Fast ICA decomposition, inverse Fast ICA, and inverse wavelet transform, and the effect of suppressing electrocardiogram interference, channel noise, system noise, and other noises is remarkable;

[0070] Step eight: the plurality of channel sEMG signals obtained in step seven are decomposed using the Fast ICA algorithm to separate the electrical activity signals of the target muscles CT muscle and LCA muscle and the surrounding interfering muscles (such as the hyothyroid muscle, the sternothyroid muscle, the thyrohyoid muscle, and the scapulohyoid muscle), and are respectively recorded as IC1, IC2, …, ICn. n wherein n is the number of channels;

[0071] Step nine: the correlation coefficients between IC i (i = 1, …, n) and the sEMG signals of the channels are calculated; and the ICs meeting the following correlation principle are found: i It is considered that the sEMG signal collected by the main sensor contains more electrical activity signals of the target muscle CT muscle of the throat, therefore, the ICs satisfying the correlation coefficient greater than 0.8 with the sEMG signal collected by the main sensor are found; iand the correlation coefficient of the sEMG signal collected by the secondary sensor is less than 0.5, record this IC as IC_target i IC_target, that is, the sEMG signal of the target muscle CT muscle of the larynx extracted by Fast ICA decomposition. If no IC i satisfying the condition is found, remove this set of multi-channel sEMG signals and do not further analyze them.

[0072] Step ten: Extract at least one time-frequency domain sEMG signal feature parameter of the sEMG signal IC_target of the target muscle CT muscle of the larynx extracted in step nine in a series of sliding rectangular analysis windows with a length of 50 ms and a step of 15 ms.

[0073] Further, in step one, the placement mode of the sensor collecting the sEMG signal of the CT muscle is mode 1, and the placement mode of the sensor collecting the sEMG signal of the LCA muscle is mode 2. Mode 1 requires that the primary sensor of at least one channel is placed on the CT muscle and parallel to the length direction of the CT muscle to collect the sEMG signal of the CT muscle; and the secondary sensor of at least one channel is placed close to the upper and lower edges of the primary sensor and parallel to the primary sensor to collect the sEMG signal of the interfering muscle around the CT muscle. Mode 2 requires that the primary sensor of at least one channel is placed on the LCA muscle and parallel to the length direction of the LCA muscle to collect the sEMG signal of the LCA muscle; and the secondary sensor of at least one channel is placed close to the upper and lower edges of the primary sensor and parallel to the primary sensor to collect the sEMG signal of the interfering muscle around the LCA muscle; by Fast ICA on the denoised signal, an equal number of independent components as the number of channels are obtained, and then the independent components that can better reflect the electrical activity of the target muscle are selected according to the correlation principle to remove the influence of the surrounding interfering muscles to the greatest extent, and at least one time-frequency domain feature of the surface electromyogram is extracted.

[0074] Further, in step one, the target muscle further includes the thyroarytenoid muscle, the sternothyroid muscle, the sternohyoid muscle, and the thyrohyoid muscle.

[0075] Further, in step three, the passband frequency of the band-pass filter is 50-350 Hz, the upper stopband frequency is 500 Hz, and the lower stopband frequency is 20 Hz.

[0076] Further, in step four, the wavelet function is Symlet4, and the decomposition scale is 8.

[0077] Further, in step ten, the time domain sEMG parameter is calculated according to the following formula:

[0078]

[0079]

[0080]

[0081]

[0082]

[0083] Wherein, s is sEMG signal, N is data length;

[0084] The frequency domain sEMG parameter is calculated according to the following formula:

[0085]

[0086]

[0087]

[0088]

[0089]

[0090]

[0091]

[0092]

[0093] Wherein, P is power spectral density, f is corresponding frequency, N is data length.

[0094] In summary, the application first collects the electrical activity signal of the laryngeal muscle through the sEMG technology and the multi-channel sensor placement method, which is a non-invasive method, can solve the disadvantage of LEMG invasiveness, and the sensor placement method provides a basis for decomposing the laryngeal target muscle electrical activity in the signal processing process, solves the problem that the existing sEMG acquisition method cannot obtain the laryngeal target muscle electrical activity signal, then through signal rectification, filtering, wavelet transform, Fast ICA decomposition, inverse FastICA and inverse wavelet, all channel surface electromyography signals are denoised, the inhibition effect of ECG interference, channel noise, system noise and other noises is remarkable, so the application has wide application prospect.

[0095] It should be emphasized that: the above is only the preferred embodiment of the application, not any form of limitation on the application, any simple modification, equivalent change and modification of the above embodiment according to the technical essence of the application still belongs to the scope of the technical solution of the application.

Claims

1. A surface myoelectric-based laryngeal muscle electrical signal acquisition and processing method, characterized in that, The method comprises the following steps: Step one: collecting sEMG signals of laryngeal muscles by using different multi-channel differential sensor placement methods, the target muscles including cricothyroid muscle and cricoarytenoid lateral muscle; Step two: rectifying the sEMG signals collected in step one into zero-mean sEMG signals; Step three: filtering out low-frequency and high-frequency noises from the zero-mean sEMG signals obtained in step two by using a Butterworth band-pass filter; Step four: performing wavelet transform on the single-channel signals obtained in step three to decompose the single-channel signals into multiple groups of signals with different frequency components; Step five: decomposing the multiple-component signals obtained in step four by using a Fast ICA algorithm to obtain the same number of independent component signals, and then selectively removing noise components, the noise components including components with ECG signals and channel noises; Step six: performing inverse Fast ICA and inverse wavelet transform on the independent component signals remaining after removing the noise components in step five to restore the multi-channel independent component signals to single-channel signals, the single-channel signals being denoised single-channel sEMG signals; Step seven: repeating steps two to six on sEMG signals of different channels until all denoised sEMG signals of the channels are obtained; Step eight: using Fast ICA algorithm to decompose the multi-channel sEMG signal obtained in step seven, separating the electrical activity signals of the target muscle cricothyroid muscle or cricothyroid lateral muscle and the surrounding interference muscles sternohyoid muscle, sternothyroid muscle, thyrohyoid muscle, and omohyoid muscle, obtaining the same number of independent components as the number of channels of the multi-channel sEMG signal, denoted as IC1, IC2, …, IC n n is the number of channels of the multi-channel sEMG signal; the independent components include the sEMG signals of the target muscle and the surrounding interference muscles; Step nine: Find IC_target in IC i i.e. the sEMG signal of the laryngeal target muscle cricothyroid or cricoarytenoid, i = 1, 2, …, n; Step ten: extracting at least one time-frequency domain sEMG signal feature parameter of the laryngeal target muscles from the sEMG signals extracted in step nine in a series of sliding rectangular analysis windows; In step nine, the IC obtained in step eight i The search for the cricothyroid muscle or cricoarytenoid muscle, the target muscles of the larynx, in the range i = 1, ..., n, is conducted by calculating the IC... i The correlation coefficient between the IC and the sEMG signals of each channel; find ICs that meet the following correlation criteria. i It is believed that the sEMG signal acquired by the main sensor contains more electrical activity signals of the target muscles in the throat. Therefore, we are looking for an IC that has a correlation coefficient greater than 0.8 with the sEMG signal acquired by the main sensor. i And satisfying the condition that the correlation coefficient between the IC and the sEMG signal acquired by the secondary sensor is less than 0.5, this IC is denoted as IC. i IC_target is the sEMG signal of the target laryngeal muscle extracted by Fast ICA decomposition; if no matching IC is found... i If so, this set of multi-channel sEMG signals is removed and no further analysis is performed.

2. The surface myoelectric-based laryngeal muscle electrical signal acquisition and processing method according to claim 1, characterized in that, In step one, when the target muscle is the cricothyroid muscle, at least one main sensor is required to be parallel to the axial direction of the cricothyroid muscle and located on the cricothyroid muscle for collecting sEMG signals of the cricothyroid muscle, and at least one auxiliary sensor is required to be parallel to the main sensor and placed close to the upper edge or lower edge of the main sensor for collecting sEMG signals of interfering muscles around the cricothyroid muscle; when the target muscle is the cricoarytenoid lateral muscle, at least one main sensor is required to be parallel to the direction of the cricoarytenoid lateral muscle and located on the cricoarytenoid lateral muscle for collecting sEMG signals of the cricoarytenoid lateral muscle, and at least one auxiliary sensor is required to be parallel to the main sensor and placed close to the upper edge or lower edge of the main sensor for collecting sEMG signals of interfering muscles around the cricoarytenoid lateral muscle.

3. The surface myoelectric-based laryngeal muscle electrical signal acquisition and processing method according to claim 1, characterized in that, In step one, the target muscles further include the thyroarytenoid muscle, sternothyroid muscle, sternothyrohyoid muscle, and thyrohyoid muscle.

4. The surface myoelectric-based laryngeal muscle electrical signal acquisition and processing method according to claim 1, characterized in that, In step three, the upper passband frequency of the band-pass filter is selected in the range of 350-500 Hz, the lower passband frequency is selected in the range of 0-50 Hz, the upper stopband frequency is 500 Hz, and the lower stopband frequency is selected in the range of 0-20 Hz.

5. The surface myoelectric-based laryngeal muscle electrical signal acquisition and processing method according to claim 1, characterized in that, In step four, the wavelet function can be selected as a Symlet wavelet or a Daubechies wavelet, and the decomposition scale is selected in the range of 6-10.

6. The surface myoelectric-based laryngeal muscle electrical signal acquisition and processing method according to claim 1, characterized in that, In step ten, the time-domain sEMG signal feature parameters include: root mean square value RMS, mean absolute value MAV, integral value INTE, variance VAR, and zero-crossing rate ZC; the frequency-domain sEMG signal feature parameters include: median frequency MDF, mean power frequency MPF, frequency ratio FR, total power TTP, 1-3 order spectral moments SM1-SM3, and central frequency variance VCF; in step ten, the time-domain sEMG parameters are calculated according to the following formula: Wherein, s is the sEMG signal, and N is the data length. The frequency-domain sEMG parameters are calculated according to the following formula: Wherein, P is the power spectral density, f is the corresponding frequency, and N is the data length.

7. The surface myoelectric-based laryngeal muscle electrical signal acquisition and processing method according to claim 1, characterized in that, In step ten, the window length of the sliding rectangular analysis window is selected in the range of 20-500 ms, and the step length can be 30%-60% of the window length.

8. The surface myoelectric-based laryngeal muscle electrical signal acquisition and processing method according to claim 7, characterized in that, In step ten, the window length of the sliding rectangular analysis window is 50 ms, and the step length is 15 ms.

Images