Muscle fatigue detection sensor signal processing method

Abstract

The invention discloses a muscle fatigue detection sensor signal processing method, and relates to the technical field of muscle fatigue detection, a CEEMDAN-MPE decomposition-CLMS dynamic filtering-entropy weight fusion closed-loop technical architecture is designed, the efficiency is effectively decomposed through a CEEMDAN-MPE joint decomposition mode, the modal aliasing limitation is broken through, and the reliability of muscle fatigue detection is improved. The CLMS closed-loop optimization and assembly line acceleration technology is combined, low time delay is guaranteed, and meanwhile smaller characteristic errors and higher signal-to-noise ratio gains are achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of muscle fatigue detection, and in particular to a method for processing sensor signals for muscle fatigue detection. Background Art

[0002] Traditional muscle fatigue detection methods mainly rely on single-signal detection, such as detecting through surface electromyogram signals (sEMG) or inertial sensor signals, and their accuracy in muscle fatigue detection is relatively low.

[0003] To this end, the Chinese utility model with the publication number CN216090526U proposes a "muscle tension detection device" for muscle fatigue detection using multiple sensors, including an elastic belt body, Velcro fasteners provided at both ends of the elastic belt body, a plurality of electromyogram sensors and a plurality of pressure sensors distributed on the inner side of the elastic belt body. The electromyogram sensors and pressure sensors are arranged at intervals along the length direction. The plurality of electromyogram sensors are respectively connected to a collection board through independent wires, and the plurality of pressure sensors are respectively connected to the collection board through independent wires. More comprehensively, a blood oxygen sensor is also installed on the elastic belt body.

[0004] During the muscle fatigue detection process, the signals collected by the sensors often contain noise and interference, which will affect subsequent signal processing and feature extraction, thereby reducing the detection accuracy. In addition, since muscle fatigue detection involves multiple sensors (such as thin-film pressure strain sensors and blood oxygen concentration sensors), how to effectively fuse the information of these sensors is also a challenge. Summary of the Invention

[0005] The present invention aims to provide a technical solution that can solve the above problems to overcome the above deficiencies.

[0006] To achieve the above object, the present invention provides the following technical solution: A method for processing sensor signals for muscle fatigue detection, the processing method comprising the following steps: S1. Decompose the signal into main and slave channels through CEEMDAN-MPE joint decomposition; screen high-entropy IMF components in the main channel, capture the noise floor in the slave channel, and achieve cross-modal synchronization through Hilbert-Huang phase alignment; S2. Implement frequency-domain constrained filtering on the IMF components through a CLMS dynamic filter, synchronously generate an aliasing degree feedback coefficient in the 0-1 interval, and dynamically adjust the filtering weight every 10 ms; S3. Implement entropy weight allocation on the data of the main and slave channels based on a spatio-temporal domain feature pool, and use a hyperbolic tangent function non-linear mapping to suppress feature saturation, and output a decision result with both classification probability distribution and continuous value prediction.

[0007] As a further solution of the present invention: Before performing step S1, a multi-modal signal matrix including pressure signals and blood oxygen signals is received in real time through a sliding window, and after being encapsulated by CRC-16 check, the processing flow is triggered.

[0008] As a further solution of the present invention: In step S1, in the setting of the CEEMDAN decomposition parameters of the CEEMDAN-MPE joint decomposition, The processing of the pressure signal is as follows: For the low-frequency pressure signal, low-amplitude white noise injection is adopted, = 0.1, the decomposition layer number m = 8, and the termination condition is that the extreme point of the residual signal ≤ 2; The processing of the blood oxygen signal is as follows: For the high-frequency blood oxygen signal, high-amplitude white noise = 0.3, the decomposition layer number m = 12, and the residual energy threshold is set to 1% of the original signal energy.

[0009] As a further solution of the present invention: In step S1, in the determination of the MPE noise component of the CEEMDAN-MPE joint decomposition, the calculation method of multi-scale permutation entropy is as follows: For each IMF component, the permutation entropy with scales s = 1 to s = 5 is calculated, and the formula is: where is the probability distribution of the symbol sequence; The screening of the noise IMF is as follows: For the pressure signal: The IMF with MPE < 0.35 is determined as noise; For the blood oxygen signal: The IMF with MPE > 0.72 is determined as motion artifact.

[0010] As a further solution of the present invention: In step S1, in the generation of the noise reference signal of the CEEMDAN-MPE joint decomposition, the screened noise IMFs are superimposed to generate a noise reference signal in the same frequency band as the original signal.

[0011] As a further solution of the present invention: In step S2, the CLMS dynamic filter is composed of a parallel connection of an LMS sub-filter and an RLS sub-filter, and the output is the weighted sum of the two: The step size of the LMS filter ; The forgetting factor of the RLS filter .

[0012] As a further solution of the present invention: In step S2, the weight iteration formula is: where is the learning rate, and e(t) is the error between the desired signal and the filtered output; The initial weight assignment is: The pressure signal is dominated by LMS, ; The blood oxygen signal is dominated by RLS 2.

[0013] As a further solution of the present invention: in step S3, the pressure signal feature extraction The pressure change rate, and the calculation formula is: As a further solution of the present invention: in step S3, the blood oxygen signal feature extraction The blood oxygen saturation decline rate, and the calculation formula is: As a further solution of the present invention: in step S3, the weight calculation method is: Dynamically adjust the fusion weight of the pressure and blood oxygen features according to the real-time signal-to-noise ratio: The definition of SNR: The comprehensive fatigue index: Compared with the prior art, the beneficial effects of the present technical solution are: by designing the above-mentioned closed-loop technical architecture of "CEEMDAN-MPE decomposition - CLMS dynamic filtering - entropy weight fusion", this architecture effectively decomposes efficiency and breaks through the mode mixing limit through the CEEMDAN-MPE joint decomposition method, and combines the CLMS closed-loop optimization and pipeline acceleration technology to achieve smaller feature errors and higher signal-to-noise ratio gain while ensuring low latency.

[0014] The additional aspects and advantages of the present invention will be partially given in the following description, partially become obvious from the following description, or be understood through the practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0016] Figure 1It is a flowchart of the method for processing sensor signals for muscle fatigue detection according to the present invention; Figure 2 It is a schematic diagram of the use of the sensor applied in the present invention. Specific embodiments

[0017] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0018] Please refer to Figure 1-2 , a method for processing sensor signals for muscle fatigue detection, the processing method includes the following steps: S1. Decompose the signal into main and slave channels through CEEMDAN-MPE joint decomposition; screen the high-entropy IMF components in the main channel (such as entropy threshold > 4.2 bit), capture the noise floor in the slave channel, and achieve cross-modal synchronization through Hilbert-Huang phase alignment; S2. Implement frequency-domain constrained filtering on the IMF components through a CLMS dynamic filter, synchronously generate an aliasing degree feedback coefficient in the 0-1 interval, and dynamically adjust the filtering weight every 10 ms; S3. Implement entropy weight allocation on the data of the main and slave channels based on the spatio-temporal domain feature pool (high-entropy channel weight ≥ 65%), use the hyperbolic tangent function non-linear mapping to suppress feature saturation, and finally output a dual-mode decision result with both classification probability distribution (Softmax) and continuous value prediction (±0.1% accuracy), In this embodiment, the input signals include a pressure signal and a blood oxygen signal.

[0019] For example, obtain the muscle pressure signal of the user through the thin-film pressure strain sensor 1, and obtain the blood oxygen signal of the user through the blood oxygen concentration sensor 2.

[0020] In some embodiments, the thin-film pressure strain sensor 2 and the blood oxygen concentration sensor 3 are installed on a holder 1, and the thin-film pressure strain sensor 2 and the blood oxygen concentration sensor 3 can be held at the corresponding parts of the user's body through the holder 1 for detection.

[0021] Both the thin-film pressure strain sensor 2 and the blood oxygen concentration sensor 3 are signal-connected to the detection device 4, such as through a wired connection method for signal connection.

[0022] Such as Figure 2 shown, the holder 1 is, for example, a sleeve that can be sleeved on the limb (such as the calf) of the user.

[0023] In some embodiments, the sleeve has a pressing ability and can be kept pressed against the user's limb when sleeved on the limb, so that the thin-film pressure strain sensor 2 and the blood oxygen concentration sensor 3 can be better close to the user's limb for detection.

[0024] The sleeve with the pressing ability is, for example, an elastic sleeve. When sleeved on the user's limb and elastically stretched by the limb, it can elastically press the limb through its elastic ability.

[0025] In some embodiments, before performing step S1, a multi-modal signal matrix including pressure signals and blood oxygen signals is received in real time through a sliding window (such as 200 ms / 30% overlap rate), and after being encapsulated by CRC-16 check, the processing flow is triggered.

[0026] In this embodiment, in step S1, the CEEMDAN-MPE joint decomposition includes CEEMDAN decomposition parameter setting, MPE noise component determination, and noise reference signal generation.

[0027] In some embodiments, in the CEEMDAN decomposition parameter setting, The processing of the pressure signal is as follows: For the low-frequency pressure signal (such as 0.1 - 5 Hz), low-amplitude white noise injection ( = 0.1) is adopted, the decomposition layer number m = 8, and the termination condition is that the extreme point of the residual signal ≤ 2.

[0028] The processing of the blood oxygen signal is as follows: For the high-frequency blood oxygen signal (such as 0.5 - 20 Hz), high-amplitude white noise ( = 0.3) is adopted, the decomposition layer number m = 12, and the residual energy threshold is set to 1% of the original signal energy.

[0029] In some embodiments, in the MPE noise component determination, The calculation method of multi-scale permutation entropy (MPE) is as follows: For each IMF component, the permutation entropy with scales s = 1 to s = 5 is calculated, and the formula is: where is the probability distribution of the symbol sequence; The screening of the noise IMF is as follows: For the pressure signal: The IMF with MPE < 0.35 is determined as noise (such as baseline drift, power frequency interference).

[0030] For the blood oxygen signal: The IMF with MPE > 0.72 is determined as motion artifacts (such as limb swing noise).

[0031] In some embodiments, in the generation of the noise reference signal, the selected noise IMFs are superimposed to generate a noise reference signal n(t) in the same frequency band as the original signal for subsequent filtering.

[0032] In this embodiment, the CLMS dynamic filter is composed of parallel LMS (Least Mean Square) and RLS (Recursive Least Square) sub-filters, and the output is the weighted sum of the two: ‌LMS filter‌: Suitable for the slow time-varying characteristics of the pressure signal, the step size .

[0033] ‌RLS filter‌: Suitable for the fast transient noise of the blood oxygen signal, the forgetting factor .

[0034] In this embodiment, in step S2, the weight iteration formula is: where is the learning rate, and e(t) is the error between the desired signal and the filtered output.

[0035] ‌The initial weight assignment is‌: The pressure signal is dominated by LMS ( ), and it has better suppression of steady-state noise.

[0036] The blood oxygen signal is dominated by RLS ( 2), and it has a faster response to burst noise.

[0037] Through simulation comparison experiments, the signal-to-noise ratio of this filter under mixed noise (Gaussian white noise + motion artifacts) is increased to 15.2 dB.

[0038] In this embodiment, in step S3, the extraction of pressure signal features ‌Pressure Change Rate (PCR): Quantifies the instantaneous intensity of muscle contraction, and the calculation formula is: In some embodiments, if the PCR continuously increases, it indicates that the muscle enters the compensatory contraction state, which can be correspondingly identified as cumulative fatigue.

[0039] In this embodiment, in step S3, the extraction of blood oxygen signal features ‌Oxygen Desaturation Rate (ODR)‌: Reflects the degree of imbalance between local tissue oxygen supply and consumption, and the calculation formula is: In some embodiments, the threshold setting: ODR < -0.5% / s for 10 seconds can be correspondingly identified as over-fatigue.

[0040] In this embodiment, in step S3, the weight calculation method is as follows: Dynamically adjust the fusion weight of pressure and blood oxygen characteristics according to the real-time signal-to-noise ratio (SNR): Definition of SNR: Comprehensive fatigue index: In this embodiment, by designing the above-mentioned closed-loop technical architecture of "CEEMDAN-MPE decomposition - CLMS dynamic filtering - entropy weight fusion", this architecture effectively decomposes efficiency (such as a 40% increase) through the CEEMDAN-MPE joint decomposition method, breaks through the mode mixing limit, combines the CLMS closed-loop optimization and pipeline acceleration technology, and realizes smaller feature errors (such as ≤ 3%) and higher signal-to-noise ratio gain (such as ≥ 15 dB) while ensuring low latency (such as ≤ 25 ms).

[0041] In some embodiments, hierarchical early warning can be performed according to the calculation results. For example: When 0.5 ≤ F(t) < 0.75, a prompt for suggesting to reduce the exercise intensity is given.

[0042] When F(t) ≥ 0.75, a prompt for immediately stopping the exercise is given to avoid muscle damage.

[0043] In some embodiments, the method of hierarchical pre-tightening is, for example, to give an interface prompt through the display screen 41 on the detection device 4, or to give a sound prompt through the speaker (not shown) on the detection device.

[0044] In some embodiments, in step S3, manual review is performed when the confidence level is lower than 90%.

[0045] For those skilled in the art, it is obvious that the present invention is not limited to the details of the above-mentioned exemplary embodiments, and the present invention can be implemented in other specific forms without departing from the spirit or basic characteristics of the present invention. Therefore, in any aspect, the embodiments should be regarded as exemplary and non-limiting. The scope of the present invention is defined by the appended claims rather than the above description. Therefore, all changes falling within the meaning and scope of the equivalent elements of the claims are intended to be encompassed within the present invention. Any reference signs in the claims should not be regarded as limiting the claims involved.

Claims

1. A muscle fatigue detection sensor signal processing method, characterized in that: The treatment method includes the following steps: S1. The signal is decomposed into master and slave channels through CEEMDAN-MPE joint decomposition. The master channel filters the high entropy IMF component, the slave channel captures the noise floor, and cross-modal synchronization is achieved through Hilbert-Huang phase alignment. S2, frequency-domain constrained filtering is performed on the IMF component through the CLMS dynamic filter, and the aliasing feedback coefficient in the range of 0-1 is generated synchronously, and the filtering weight is dynamically adjusted every 10ms; S3. Based on the spatiotemporal feature pool, entropy weight allocation is implemented for the master and slave channel data, and the hyperbolic tangent function nonlinear mapping is used to suppress feature saturation, and the decision result with both classification probability distribution and continuous value prediction is output.

2. The muscle fatigue detection sensor signal processing method according to claim 1, characterized in that: Before performing step S1, The multimodal signal matrix including pressure signal and blood oxygen signal is received in real time through the sliding window, and the processing flow is triggered after CRC-16 verification and packaging.

3. The muscle fatigue detection sensor signal processing method according to claim 1 or 2, characterized in that: In step S1, in the CEEMDAN decomposition parameter setting of the CEEMDAN-MPE joint decomposition, The pressure signal is processed as follows: The low-frequency pressure signal is injected with low-amplitude white noise. =0.1, the number of decomposition layers m=8, and the termination condition is that the extreme value of the residual signal is ≤2; The blood oxygen signal is processed as follows: High-frequency blood oxygen signal uses high-amplitude white noise =0.3, the number of decomposition layers m=12, and the residual energy threshold is set to 1% of the original signal energy.

4. The muscle fatigue detection sensor signal processing method according to claim 3, characterized in that: In step S1, in the determination of the MPE noise component of the CEEMDAN-MPE joint decomposition, the multi-scale permutation entropy is calculated as: For each IMF component, the permutation entropy of scale s=1 to s=5 is calculated as follows: in is the probability distribution of symbol sequence; ‌Noise IMF filter is‌: ‌Pressure signal‌: IMF with MPE < 0.35 is considered as noise; ‌Blood oxygen signal‌: IMF with MPE > 0.72 is considered a motion artifact.

5. The muscle fatigue detection sensor signal processing method according to claim 4, characterized in that: In step S1, in the noise reference signal generation of the CEEMDAN-MPE joint decomposition, the screened noise IMFs are superimposed to generate a noise reference signal in the same frequency band as the original signal.

6. The muscle fatigue detection sensor signal processing method according to claim 1, 2, 4 or 5, characterized in that: In step S2, the CLMS dynamic filter is composed of LMS and RLS sub-filters in parallel, and the output is the weighted sum of the two: ‌LMS filter step size ; ‌RLS filter forgetting factor .

7. The muscle fatigue detection sensor signal processing method according to claim 6, characterized in that: In step S2, the weight iteration formula is: in is the learning rate, e(t) is the error between the expected signal and the filtered output; The initial weight distribution is: The pressure signal is dominated by LMS. ; Blood oxygen signal is dominated by RLS 2.

8. The muscle fatigue detection sensor signal processing method according to claim 1 or 2 or 4 or 5 or 7, characterized in that: In step S3, the pressure signal feature extraction ‌Pressure change rate, calculated as:

9. The muscle fatigue detection sensor signal processing method according to claim 8, characterized in that: In step S3, blood oxygen signal feature extraction ‌ Blood oxygen saturation decrease rate, calculated as:

10. The muscle fatigue detection sensor signal processing method according to claim 9, characterized in that: In step S3, the weight is calculated as follows: Dynamically adjust the fusion weight of pressure and blood oxygen characteristics according to the real-time signal-to-noise ratio: ‌SNR Definition‌: ‌Comprehensive Fatigue Index‌:

Images