Motion state detection method and device based on multi-modal sensor

Abstract

The invention relates to a motion state detection method and device based on a multi-modal sensor, and the method comprises the following steps: carrying out the inertial parameter extraction of an original acceleration signal in the motion process of a human body through a three-axis acceleration sensor, and obtaining a motion acceleration feature vector; the original pulse wave signals in the movement process are monitored in real time, and a cardiovascular physiological feature sequence is obtained; performing time-frequency domain conjoint analysis on the time sequence myoelectricity feature data set, the motion acceleration feature vector and the cardiovascular physiological feature sequence by adopting an adaptive wavelet transform technology to obtain a multi-dimensional feature fusion matrix; and performing dynamic segmentation on the multi-dimensional feature fusion matrix through a hierarchical clustering algorithm to obtain motion state feature subspaces, thereby solving the problem that a traditional motion monitoring method mainly depends on a single type of sensor, although the motion condition of a human body can be reflected to a certain degree, the motion state feature subspaces cannot be monitored. However, the technical problem of lack of comprehensive understanding of complex motion states is solved.

Description

Technical Field

[0001] The present invention relates to the field of sensor technology, and in particular to a motion state detection method and device based on a multi-modal sensor. Background Art

[0002] In modern society, with the improvement of people's health awareness and the popularization of fitness culture, the demand for accurate monitoring of exercise status is increasing. Traditional exercise monitoring methods mainly rely on a single type of sensor, such as accelerometers or heart rate monitors. Although these methods can reflect the movement of the human body to a certain extent, they often lack a comprehensive understanding of complex movement states. For example, when evaluating the performance of athletes or the effectiveness of rehabilitation training, relying solely on simple exercise data cannot provide enough information to accurately judge the body's response and recovery status.

[0003] In addition, although the application of multimodal sensors in existing technologies has been developed, it still faces many challenges in practical applications. On the one hand, the types of data collected by different types of sensors (such as bioelectric sensors, acceleration sensors, and photoelectric volume pulse wave sensors) are quite different, and how to effectively integrate these heterogeneous data has become a difficult problem; on the other hand, due to the dynamics and complexity of human motion, how to achieve efficient and real-time data collection and analysis without affecting the user's freedom of movement is also an urgent problem to be solved. These problems limit the application and development of motion state detection technology based on multimodal sensors in a wider range of fields.

[0004] Faced with the above challenges, researchers are committed to exploring more comprehensive and accurate motion state detection solutions. A key direction is to develop algorithms and technologies that can process multiple types of physiological signals simultaneously to achieve a more detailed and in-depth understanding of the human body's motion state. By fusing data from different sensors and using advanced signal processing technology for in-depth analysis, subtle changes in the body during exercise can be better captured. This not only helps to improve athletic performance and prevent sports injuries, but also provides new possibilities for personalized health management. However, in order for this technology to truly mature and be widely used, many technical and practical obstacles need to be overcome. Summary of the invention

[0005] The main purpose of the present invention is to provide a motion state detection method and device based on a multimodal sensor, which solves the technical problem that traditional motion monitoring methods mainly rely on a single type of sensor, which can reflect the movement of the human body to a certain extent, but often lack a comprehensive understanding of complex motion states.

[0006] To achieve the above object, the present invention provides a motion state detection method based on a multimodal sensor, which is applied to a motion state detection device, wherein the motion state detection device includes a bioelectric sensor array, a three-axis acceleration sensor and a photoelectric volume pulse wave sensor, and includes the following steps: The bioelectric sensor array is used to collect the electromyographic signals at multiple points during human motion to obtain a time-series electromyographic feature data set. The inertial parameters of the original acceleration signal during human motion are extracted through a three-axis acceleration sensor to obtain a motion acceleration feature vector; Based on the photoplethysmography sensor, the original pulse wave signal during exercise is monitored in real time to obtain the cardiovascular physiological characteristic sequence; Adopting adaptive wavelet transform technology to perform time-frequency domain joint analysis on the time-series electromyographic feature data set, the motion acceleration feature vector and the cardiovascular physiological feature sequence to obtain a multi-dimensional feature fusion matrix; The multi-dimensional feature fusion matrix is ​​dynamically segmented through a hierarchical clustering algorithm to obtain a motion state feature subspace.

[0007] Furthermore, the bioelectric sensor array is provided with a differential amplifier circuit, and the bioelectric sensor array is used to collect multi-point electromyographic signals during human body movement to obtain a time-series electromyographic characteristic data set, including: The electromyographic signals during human motion are collected by a bioelectric sensor array, and the electromyographic signals are pre-amplified and common-mode suppressed by a differential amplifier circuit to obtain a multi-channel electromyographic signal sequence; Performing instantaneous frequency analysis and nonlinear demodulation on the multi-channel electromyographic signal sequence to obtain an electromyographic signal modulation feature set; The second-order moment analysis method is used to extract time-varying features and calculate spatial correlation of the EMG signal modulation feature set to obtain a time-series EMG feature data set; wherein the time-series EMG feature data set includes motor unit action potential, muscle fiber conduction velocity and muscle fatigue data.

[0008] Furthermore, the inertial parameter extraction of the original acceleration signal during the human body movement by the three-axis acceleration sensor to obtain the motion acceleration feature vector includes: The original acceleration signal during human motion is collected by the three-axis acceleration sensor, and the original acceleration signal is subjected to posture compensation and noise elimination by the Kalman compensator to obtain a compensated acceleration data stream, and the gravity component of the compensated acceleration data stream is separated to obtain a pure inertial acceleration sequence; wherein the pure inertial acceleration sequence includes a vertical motion component, a horizontal motion component and a rotational motion component; The pure inertial acceleration sequence is subjected to coordinate system conversion and posture calculation by a preset quaternion transformation technology to obtain a human motion posture parameter set, and a motion trajectory is reconstructed based on the human motion posture parameter set to obtain a three-dimensional space motion trajectory sequence; wherein the three-dimensional space motion trajectory sequence includes a position coordinate sequence, a velocity vector sequence and an acceleration vector sequence; Based on a multi-scale morphological analysis method, feature extraction is performed on the three-dimensional space motion trajectory sequence to obtain a motion morphology feature group, and the motion morphology feature group is transformed into a spectrum domain to obtain a motion spectrum feature sequence; wherein the motion spectrum feature sequence includes a main frequency component, a harmonic component and a modulation component; The motion frequency spectrum feature sequence is subjected to dimension reduction and feature fusion through singular value decomposition to obtain a motion acceleration feature vector.

[0009] Furthermore, the photoplethysmography sensor is used to monitor the original pulse wave signal during exercise in real time to obtain a cardiovascular physiological characteristic sequence, including: The original pulse wave signal during the movement is collected by a photoelectric volumetric pulse wave sensor, and the morphological features of the original pulse wave signal are extracted to obtain a pulse wave morphological feature set; wherein the pulse wave morphological feature set includes a pulse wave peak value, a double peak interval, and a trough elasticity coefficient; Based on wavelet entropy analysis, the pulse wave morphological feature set is dynamically decomposed in the time-frequency domain to obtain a pulse wave spectrum component array, and the pulse wave spectrum component array is subjected to spectrum peak tracking and phase difference accumulation to obtain a heart rate variability data group; wherein the heart rate variability data group includes low-frequency power density, high-frequency power density ratio and spectrum entropy; Performing a long-term correlation analysis on the heart rate variability data set by using a Hurst index calculation method to obtain a cardiac autonomic nerve regulation parameter set, and dynamically estimating blood pressure based on the cardiac autonomic nerve regulation parameter set to obtain a continuous blood pressure change curve; wherein the continuous blood pressure change curve includes systolic pressure, diastolic pressure and mean arterial pressure; The vascular compliance of the continuous blood pressure change curve is evaluated based on a preset fractional-order derivative model to obtain a peripheral vascular resistance index, and a multivariate coupling analysis is performed on the peripheral vascular resistance index and the heart rate variability data group to obtain a cardiovascular physiological characteristic sequence; wherein the cardiovascular physiological characteristic sequence includes a cardiac load index, a vascular regulation responsiveness, and an autonomic nervous balance coefficient.

[0010] Furthermore, the vascular compliance evaluation of the continuous blood pressure change curve based on the preset fractional derivative model to obtain the peripheral vascular resistance index includes: The continuous blood pressure change curve is subjected to non-integer derivative decomposition by using a Caputo fractional differential operator to obtain a vascular wall stress-strain relationship curve, and the vascular wall stress-strain relationship curve is subjected to piecewise polynomial fitting to obtain a vascular elastic modulus distribution function; wherein the vascular elastic modulus distribution function includes a large artery compliance parameter, a microcirculatory resistance coefficient, and a vascular wall viscoelastic factor; Based on the preset Windkessel pulsating blood flow model, the vascular elastic modulus distribution function is coupled with hemodynamic parameters to obtain a vascular pressure-volume response characteristic spectrum, and the vascular pressure-volume response characteristic spectrum is subjected to nonlinear regression analysis to obtain a vascular compliance attenuation curve; wherein the vascular compliance attenuation curve includes an exponential attenuation parameter, a time constant matrix, and an elastic recovery coefficient; The energy distribution of the vascular compliance attenuation curve is extracted in the time-frequency domain by using a wavelet packet entropy value calculation method to obtain a vascular function state set, and the pulse pressure conduction velocity is estimated based on the vascular function state set to obtain an arteriosclerosis index; wherein the arteriosclerosis index includes a pulse pressure amplification factor, a reflection wave enhancement factor, and a pulse conduction delay time; The vascular-cardiac feedback mechanism of the arteriosclerosis index and the heart rate variability data group is modeled by a preset fractional derivative model to obtain a peripheral vascular resistance index; wherein the peripheral vascular resistance index includes a vascular tension regulation coefficient, a sympathetic-parasympathetic balance ratio and a vascular reactivity reserve value.

[0011] Furthermore, the adaptive wavelet transform technology is used to perform a time-frequency domain joint analysis on the time-series electromyographic feature data set, the motion acceleration feature vector and the cardiovascular physiological feature sequence to obtain a multi-dimensional feature fusion matrix, including: Adopting adaptive wavelet transform technology to perform time-frequency decomposition on the time-series electromyographic feature data set to obtain an electromyographic signal wavelet coefficient matrix, and performing energy distribution calculation on the electromyographic signal wavelet coefficient matrix to obtain an electromyographic activity energy density spectrum; Performing multi-scale decomposition on the motion acceleration feature vector by using adaptive wavelet transform technology to obtain a motion acceleration wavelet coefficient array, and performing singular spectrum analysis on the motion acceleration wavelet coefficient array to obtain a motion dynamics spectrum map; Based on the adaptive wavelet transform technology, a time-varying spectrum analysis is performed on the cardiovascular physiological characteristic sequence to obtain a cardiovascular response wavelet coefficient group, and a bispectral coherence estimation is performed on the cardiovascular response wavelet coefficient group to obtain a cardiopulmonary coupling characteristic spectrum; The electromyographic activity energy density map, the motion dynamics spectrum map and the cardiopulmonary coupling feature map are aligned in time series by Hilbert-Huang transform to obtain a multimodal feature synchronization matrix, and the multimodal feature synchronization matrix is ​​subjected to high-order tensor decomposition to obtain a multidimensional feature fusion matrix; wherein the multidimensional feature fusion matrix includes motion-cardiovascular coupling mode, electromyographic-acceleration synergy parameters and physiological response integration feature space.

[0012] Furthermore, the multi-dimensional feature fusion matrix is ​​dynamically segmented by a hierarchical clustering algorithm to obtain a motion state feature subspace, including: Performing density peak detection on the multi-dimensional feature fusion matrix through adaptive threshold segmentation to obtain a feature density topological distribution map, and performing kernel density contour extraction on the feature density topological distribution map to obtain a multi-dimensional feature clustering boundary set, wherein the multi-dimensional feature clustering boundary set includes an exercise intensity boundary, a cardiovascular response boundary, and an electromyographic activity boundary; Based on the minimum spanning tree technology, hierarchical distance calculation is performed on the multidimensional feature clustering boundary set to obtain a feature space hierarchical structure, and the feature space hierarchical structure is adaptively truncated to obtain a dynamic hierarchical tree structure, wherein the dynamic hierarchical tree structure includes main motion mode nodes, transition state nodes and abnormal state nodes; The dynamic hierarchical tree structure is subjected to Laplace matrix decomposition by a spectral clustering method to obtain a characteristic subspace orthogonal basis, and the characteristic subspace orthogonal basis is subjected to manifold consistency evaluation to obtain a state classification judgment boundary; wherein the state classification judgment boundary includes a steady-state motion region, an acceleration transition region, and a fatigue deceleration region; Based on a preset kernel principal component analysis method, the state classification discrimination boundary is nonlinearly mapped to obtain a motion state feature projection matrix, and high-order statistics are extracted from the motion state feature projection matrix through a hierarchical clustering algorithm to obtain a motion state feature subspace; wherein the motion state feature subspace includes a static state feature set, a uniform motion feature set, a variable speed motion feature set and a fatigue state feature set.

[0013] The present invention also provides a motion state detection device based on a multimodal sensor, which is applied to a motion state detection device. The motion state detection device includes a bioelectric sensor array, a three-axis acceleration sensor and a photoelectric volume pulse wave sensor, including: The acquisition module is used to use a bioelectric sensor array to perform multi-point acquisition of electromyographic signals during human motion to obtain a time-series electromyographic feature data set; An extraction module is used to extract inertial parameters from the original acceleration signal during human motion using a three-axis acceleration sensor to obtain a motion acceleration feature vector; A monitoring module, which is used to monitor the original pulse wave signal during exercise in real time based on a photoplethysmographic sensor to obtain a cardiovascular physiological characteristic sequence; An analysis module, for performing a time-frequency domain joint analysis on the time-series electromyographic feature data set, the motion acceleration feature vector and the cardiovascular physiological feature sequence using an adaptive wavelet transform technique to obtain a multi-dimensional feature fusion matrix; The segmentation module is used to dynamically segment the multi-dimensional feature fusion matrix through a hierarchical clustering algorithm to obtain a motion state feature subspace.

[0014] The present invention also provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of any one of the above methods when executing the computer program.

[0015] The present invention also provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of any of the above-mentioned methods are implemented.

[0016] The present invention provides a motion state detection method based on a multimodal sensor, comprising the following steps: using a bioelectric sensor array to perform multi-point acquisition of electromyographic signals during human motion to obtain a time-series electromyographic feature data set; using a three-axis acceleration sensor to extract inertial parameters of the original acceleration signal during human motion to obtain a motion acceleration feature vector; based on a photoelectric volumetric pulse wave sensor, real-time monitoring of the original pulse wave signal during motion to obtain a cardiovascular physiological feature sequence; using an adaptive wavelet transform technology to perform a time-frequency domain joint analysis of the time-series electromyographic feature data set, the motion acceleration feature vector and the cardiovascular physiological feature sequence to obtain a multi-dimensional feature fusion matrix; using a hierarchical clustering algorithm to dynamically segment the multi-dimensional feature fusion matrix to obtain a motion state feature subspace. Through the above technical means, the technical problem that the traditional motion monitoring method mainly relies on a single type of sensor and can reflect the motion of the human body to a certain extent but often lacks a comprehensive understanding of the complex motion state is solved, and the multi-dimensional feature fusion matrix is ​​dynamically segmented by a hierarchical clustering algorithm to form a motion state feature subspace. This process can identify and distinguish different movement patterns or states, providing strong support for personalized exercise monitoring and health management. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 is a schematic diagram of the steps of a motion state detection method based on a multimodal sensor in one embodiment of the present invention; Figure 2 is a structural block diagram of a motion state detection device based on a multi-modal sensor in one embodiment of the present invention; Figure 3 It is a schematic block diagram of the structure of a computer device according to an embodiment of the present invention.

[0018] The purpose, features and advantages of the present invention will be further described with reference to the accompanying drawings in conjunction with the embodiments. DETAILED DESCRIPTION

[0019] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0020] like Figure 1 As shown, Figure 1 It is a schematic diagram of the steps of a motion state detection method based on a multimodal sensor in one embodiment of the present invention; In one embodiment of the present invention, a motion state detection method based on a multimodal sensor is provided, which is applied to a motion state detection device, wherein the motion state detection device includes a bioelectric sensor array, a three-axis acceleration sensor, and a photoelectric volume pulse wave sensor, and includes the following steps: Step S1, using a bioelectric sensor array to perform multi-point acquisition of electromyographic signals during human motion to obtain a time-series electromyographic feature data set.

[0021] Specifically, a bioelectric sensor array is used to collect the electromyographic signals during human movement at multiple points to obtain a time-series electromyographic feature data set. This process first relies on a carefully arranged bioelectric sensor array, which is designed to capture the weak electrical signals generated by muscle activity during human movement. By distributing multiple sensors on key muscle groups of the human body, multi-point synchronous collection of electromyographic signals can be achieved, thereby obtaining more comprehensive and accurate data. Specifically, each sensor is responsible for monitoring the changes in electromyographic signals in a specific area. When the human body performs various movements, such as running, jumping, or weightlifting, these sensors will record the corresponding electrical signals in real time and convert them into digital signals for subsequent processing. For example, in an athlete training scenario, a coach can analyze these time-series electromyographic feature data sets to evaluate whether the athlete's technical movements are correct, whether there is a risk of overusing certain muscle groups, or whether the training intensity needs to be adjusted. Furthermore, in order to ensure that the collected data has high reliability and accuracy, the layout of the sensors must be optimized so that they can cover all key muscle parts while avoiding mutual interference between signals. In addition, due to the differences in body shape and muscle distribution among different individuals, the position and number of sensors need to be adjusted according to the specific situation of the individual in practical applications. After such a detailed configuration, the bioelectric sensor array can effectively capture the subtle changes in muscle activity during exercise, thereby forming a detailed time-series electromyographic feature data set. This set of data not only reflects the degree of activation and working mode of muscles in different states of motion, but also lays the foundation for the subsequent joint analysis of time and frequency domains using adaptive wavelet transform technology, which helps to gain a deeper understanding of the human movement mechanism and guide the design of personalized training programs. For example, in rehabilitation treatment, doctors can monitor the patient's recovery progress based on these data and adjust the treatment strategy in time to promote faster and better rehabilitation effects. In this way, the application of bioelectric sensor arrays not only improves the accuracy of motion monitoring, but also provides strong support for the research of health management and sports science.

[0022] Step S2, extracting inertial parameters from the original acceleration signal during human motion using a three-axis acceleration sensor to obtain a motion acceleration feature vector.

[0023] Specifically, the inertial parameters of the original acceleration signal during human motion are extracted by a three-axis acceleration sensor to obtain the motion acceleration feature vector. This process first relies on high-precision three-axis acceleration sensors, which can capture the linear acceleration changes of the human body in three-dimensional space. When the human body performs various movements, such as running, jumping or weightlifting, the three-axis acceleration sensor will record the acceleration data in all directions in real time, including the front and back, left and right, and up and down dimensions. These original acceleration signals contain rich information and reflect the dynamic characteristics of the human body during motion. In order to extract meaningful inertial parameters from these complex signals, a series of advanced signal processing technologies are usually required. Specifically, after obtaining the original acceleration signal, the next key step is to extract the inertial parameters. This step involves filtering, denoising, and feature extraction of the original signal in order to separate the features closely related to the motion state. For example, the velocity and displacement of the human body can be determined by calculating the rate of change of the acceleration signal, and the acceleration characteristics under different motion modes can be further analyzed. The purpose of this is to construct a feature vector that can fully describe the human motion state. This vector not only contains basic acceleration information, but also integrates other important inertial parameters such as angular velocity and posture angle. Taking athlete training as an example, coaches can use these motion acceleration feature vectors to evaluate whether the athlete's technical movements are standardized, whether there is a potential risk of injury, or whether the expected training effect has been achieved. For example, in the long jump event, by analyzing the acceleration feature vector at the moment of take-off, the athlete's force and take-off angle can be accurately judged, thereby providing a scientific basis for optimizing the training plan. In addition, for rehabilitation treatment scenarios, doctors can also use these feature vectors to monitor the patient's recovery progress and ensure the effectiveness and safety of rehabilitation training. In this way, the application of triaxial acceleration sensors not only improves the accuracy of human motion state monitoring, but also provides strong support for the design of personalized training programs and the management of health conditions. Ultimately, these detailed motion acceleration feature vectors lay a solid foundation for further data analysis and algorithm development, making the motion state detection method based on multimodal sensors more complete and efficient.

[0024] Step S3, real-time monitoring of the original pulse wave signal during exercise based on the photoplethysmography sensor to obtain a cardiovascular physiological characteristic sequence.

[0025] Specifically, the original pulse wave signal during exercise is monitored in real time based on the photoelectric volumetric pulse wave sensor to obtain the cardiovascular physiological feature sequence. This process first relies on the high sensitivity and real-time performance of the photoelectric volumetric pulse wave sensor. This sensor measures blood flow by emitting light of a specific wavelength to the surface of the skin and detecting the changes in the intensity of the reflected or transmitted light. During human exercise, the sensor continuously records the changes in these light intensities to obtain the original pulse wave signal. These signals contain rich information, such as cardiovascular physiological parameters such as heart rate, blood oxygen saturation, and vascular elasticity. In order to extract meaningful cardiovascular physiological features from these complex original signals, a series of advanced signal processing technologies are usually required. Specifically, after obtaining the original pulse wave signal, the next key step is to analyze and process it in real time to extract a feature sequence that can reflect the cardiovascular health status. This includes filtering to remove noise interference, identifying the main feature points of the pulse waveform (such as peaks and troughs), and determining important indicators such as heart rate variability by calculating the time intervals of these feature points. In addition, algorithms can be used to further analyze advanced parameters such as pulse wave conduction velocity to comprehensively evaluate the state of the cardiovascular system. For example, in the athlete training scenario, the coach can monitor the athlete's physical load and recovery by analyzing these cardiovascular physiological feature sequences to ensure that the training plan can achieve the expected results without causing excessive fatigue or injury risks to the athlete. For example, during marathon training, athletes may experience long periods of high-intensity exercise. At this time, the photoelectric volume pulse wave sensor can monitor their heart rate and blood oxygen saturation changes in real time. If the heart rate is found to be continuously too high or the blood oxygen saturation drops significantly, it means that the athlete may be in a state of excessive fatigue and needs to adjust the training intensity or arrange rest in time. Similarly, in the rehabilitation treatment scenario, doctors can also use this data to monitor the patient's recovery progress to ensure the effectiveness and safety of rehabilitation training. In this way, the application of photoelectric volume pulse wave sensors not only improves the accuracy of human cardiovascular status monitoring, but also provides strong support for the design of personalized training programs and health management. Ultimately, these detailed cardiovascular physiological feature sequences lay a solid foundation for further data analysis and algorithm development, making the motion state detection method based on multimodal sensors more complete and efficient.

[0026] Step S4, using adaptive wavelet transform technology to perform time-frequency domain joint analysis on the time-series electromyographic feature data set, the motion acceleration feature vector and the cardiovascular physiological feature sequence to obtain a multi-dimensional feature fusion matrix.

[0027] Specifically, the adaptive wavelet transform technology is used to perform a joint analysis of the time-series electromyographic feature data set, the motion acceleration feature vector and the cardiovascular physiological feature sequence in the time and frequency domain to obtain a multi-dimensional feature fusion matrix. This process first relies on the powerful signal processing capability of the adaptive wavelet transform technology. The adaptive wavelet transform can automatically adjust its basis function according to the characteristics of the input signal, thereby providing high-resolution analysis results in both the time domain and the frequency domain. Specifically, after obtaining the time-series electromyographic feature data set, the motion acceleration feature vector and the cardiovascular physiological feature sequence, these different types of data will be sent to the adaptive wavelet transform algorithm for processing. In this way, not only can the key features of each signal be effectively extracted, but also their changing laws on different time scales can be captured. In actual operation, the adaptive wavelet transform first decomposes each type of raw data and converts it into a series of wavelet coefficients. These coefficients contain information about the signal in different frequency bands, so that subsequent analysis can understand the changing trend of each feature in more detail. Next, these wavelet coefficients are used to construct a multi-dimensional feature fusion matrix, which not only integrates information from electromyographic signals, acceleration signals, and pulse wave signals, but also reveals the intrinsic connection between them through joint analysis. For example, in the athlete training scenario, the coach can analyze this multi-dimensional feature fusion matrix to comprehensively evaluate whether the athlete's technical movements are correct, whether there is a risk of overusing certain muscle groups, or whether the training intensity needs to be adjusted to optimize cardiopulmonary function. For example, in the long jump event, the athlete's muscle activity, body posture changes, and cardiovascular reactions at the moment of take-off are all key factors affecting performance. By analyzing these data using adaptive wavelet transform technology, the working mode of the main force-generating muscles at take-off can be identified from the time-series electromyographic feature data set; the body acceleration changes at the moment of force generation can be extracted from the motion acceleration feature vector; and the heart rate acceleration and its recovery rate can be obtained from the cardiovascular physiological feature sequence. Together, these information constitute a detailed multi-dimensional feature fusion matrix to help coaches fully understand the athlete's performance and formulate a more scientific and reasonable training plan based on this. In addition, in the rehabilitation treatment scenario, doctors can also use these data to monitor the patient's recovery progress and ensure the effectiveness and safety of rehabilitation training. In this way, the application of adaptive wavelet transform technology not only improves the accuracy of human motion state monitoring, but also provides strong support for the design of personalized training programs and health management. Ultimately, these detailed multi-dimensional feature fusion matrices lay a solid foundation for further data analysis and algorithm development, making the motion state detection method based on multimodal sensors more complete and efficient.

[0028] Step S5, dynamically segmenting the multi-dimensional feature fusion matrix through a hierarchical clustering algorithm to obtain a motion state feature subspace.

[0029] Specifically, the multi-dimensional feature fusion matrix is ​​dynamically segmented by a hierarchical clustering algorithm to obtain a motion state feature subspace. This process first relies on the powerful data classification ability of the hierarchical clustering algorithm. After obtaining the multi-dimensional feature fusion matrix, the comprehensive information containing electromyographic signals, acceleration signals and cardiovascular physiological characteristics will be further analyzed to reveal the intrinsic connection between different motion states. The hierarchical clustering algorithm calculates the similarity or distance between data points, gradually merges the most similar data points or clusters, and forms a tree structure (i.e., a dendrogram), thereby realizing the dynamic segmentation of data. Specifically, when applying the hierarchical clustering algorithm, it is first necessary to define a suitable distance measurement method to measure the similarity between each feature vector. Commonly used measurement methods include Euclidean distance, Manhattan distance, etc. Based on the selected distance measurement method, the algorithm calculates the distance between each two feature vectors and starts to build a dendrogram based on these distance values. With the continuous iteration of the algorithm, similar feature vectors will be gradually classified into one category, and finally multiple representative feature subspaces will be formed. Each feature subspace corresponds to a specific motion state or mode, such as running, jumping, resting and other different activity types. Taking the athlete training scenario as an example, coaches can comprehensively evaluate the athlete's performance and develop personalized training plans by analyzing these motion state feature subspaces. For example, during marathon training, the athlete's physical state will undergo many changes, from high-intensity sprints to low-intensity recovery runs, and then to different stages in interval training. By dynamically segmenting the multi-dimensional feature fusion matrix through a hierarchical clustering algorithm, the feature subspaces under these different motion states can be identified, thereby helping coaches better understand the athlete's physical reactions and performance at different training stages. If it is found that a certain feature subspace reflects an abnormal increase in the athlete's heart rate or a degree of muscle fatigue that exceeds expectations, the coach can adjust the training intensity or arrange appropriate rest time in a timely manner based on this information. For another example, in the rehabilitation treatment scenario, doctors can also use this method to monitor the patient's recovery progress. By analyzing the multi-dimensional feature fusion matrix of the patient in rehabilitation training, different motion state feature subspaces can be dynamically segmented to understand the changes in the patient's physical condition at different stages of rehabilitation. For example, by observing the cardiovascular health indicators reflected in a certain feature subspace, it can be determined whether the patient has the ability to perform higher intensity training. In this way, the application of hierarchical clustering algorithm not only improves the accuracy of human motion state monitoring, but also provides strong support for the design of personalized training programs and health management. Ultimately, these detailed motion state feature subspaces lay a solid foundation for further data analysis and algorithm development, making the motion state detection method based on multimodal sensors more complete and efficient.

[0030] In a specific embodiment, the bioelectric sensor array is provided with a differential amplifier circuit, and the bioelectric sensor array is used to collect multi-point electromyographic signals during human body movement to obtain a time-series electromyographic characteristic data set, including: The electromyographic signals during human motion are collected by a bioelectric sensor array, and the electromyographic signals are pre-amplified and common-mode suppressed by a differential amplifier circuit to obtain a multi-channel electromyographic signal sequence; Performing instantaneous frequency analysis and nonlinear demodulation on the multi-channel electromyographic signal sequence to obtain an electromyographic signal modulation feature set; The second-order moment analysis method is used to extract time-varying features and calculate spatial correlation of the EMG signal modulation feature set to obtain a time-series EMG feature data set; wherein the time-series EMG feature data set includes motor unit action potential, muscle fiber conduction velocity and muscle fatigue data.

[0031] Specifically, the bioelectric sensor array is provided with a differential amplifier circuit, and the bioelectric sensor array is used to collect the electromyographic signals during human body movement at multiple points to obtain a time-series electromyographic feature data set. This process first relies on a carefully designed bioelectric sensor array and its built-in differential amplifier circuit. In actual operation, the bioelectric sensor array is arranged on the key muscle groups of the human body to capture the electromyographic signals in real time during movement. These sensors can not only sense the changes in weak electrical signals, but also pre-amplify and common-mode suppress these signals through the differential amplifier circuit, thereby improving the quality and signal-to-noise ratio of the signal. Specifically, the differential amplifier circuit amplifies the voltage difference between the two input terminals and effectively suppresses the common-mode noise, thereby ensuring that the collected multi-channel electromyographic signal sequence has high accuracy and reliability. Next, after obtaining a high-quality multi-channel electromyographic signal sequence, the further analysis step is to perform instantaneous frequency analysis and nonlinear demodulation to extract the modulation feature set of the electromyographic signal. Instantaneous frequency analysis is a method that can reflect the frequency characteristics of a signal changing over time, while nonlinear demodulation can reveal the complex dynamic characteristics hidden inside the signal. By combining these two techniques, rich modulation features can be extracted from the original EMG signals, such as changes in frequency components and fluctuations in signal envelopes. These modulation feature sets not only contain basic information about muscle activity, but also reflect changes in the activation patterns and conduction velocity of muscle fibers under different motion states. Subsequently, in order to further explore the temporal and spatial information contained in these modulation feature sets, the second-order moment analysis method is used to extract time-varying features and calculate spatial correlation of the EMG signal modulation feature set. The second-order moment analysis method can effectively extract the time-varying features of the signal and calculate the spatial correlation between signals from different channels by analyzing the statistical characteristics of the signal. This analysis can not only reveal the dynamic changes of the EMG signal in time, but also show the coordination and synchronization between different muscle groups in space. Finally, through this series of complex processing steps, a time-series EMG feature data set containing rich information is obtained. This data set includes key data such as motor unit action potential, muscle fiber conduction velocity, and muscle fatigue, providing a solid foundation for subsequent movement state evaluation. For example, in an athlete training scenario, a coach can analyze these time-series electromyographic feature data sets to fully understand whether the athlete's technical movements are correct, whether there is a risk of overusing certain muscle groups, or whether the training intensity needs to be adjusted. For example, in the long jump event, the action potential of the main force-generating muscle group at the moment of take-off, the muscle fiber conduction velocity, and muscle fatigue are all important factors affecting the performance. The time-series electromyographic feature data set processed by the above steps can record the changes in these key parameters in detail. If it is found that the conduction velocity of a muscle group is significantly reduced or the fatigue level increases rapidly, it means that the muscle group may be close to its limit load. The coach can adjust the training plan in time based on this information to avoid injuries to the athlete.For example, in the rehabilitation treatment scenario, doctors can also use this method to monitor the patient's recovery progress. By analyzing the patient's time-series electromyographic feature data set during rehabilitation training, the recovery of muscle function can be dynamically monitored. For example, by observing the changes in the conduction velocity and fatigue of a muscle group, it can be determined whether the patient has the ability to perform higher-intensity training. If it is found that the patient's conduction velocity gradually returns to normal and the fatigue level remains at a low level over a period of time, it indicates that the function of the muscle group is gradually recovering, and doctors can formulate a more scientific and reasonable rehabilitation plan based on this. In this way, the bioelectric sensor array and its supporting data processing method not only improve the accuracy of human motion state monitoring, but also provide strong support for the design of personalized training programs and health management. Ultimately, these detailed time-series electromyographic feature data sets lay a solid foundation for further data analysis and algorithm development, making the motion state detection method based on multimodal sensors more complete and efficient. In addition, considering individual differences, the specific conditions of different athletes or patients may be different, so in practical applications, the sensor layout and data analysis methods need to be appropriately adjusted according to individual characteristics. For example, for larger athletes, the number of sensors may need to be increased to cover more muscle groups, while for patients in recovery, the recovery of specific muscle groups may need to be monitored more carefully. By flexibly applying these advanced technologies and methods, the needs of different application scenarios can be better met, and more accurate and personalized sports status monitoring and management can be achieved.

[0032] In a specific embodiment, the inertial parameter extraction of the original acceleration signal during human motion by a three-axis acceleration sensor to obtain a motion acceleration feature vector includes: The original acceleration signal during human motion is collected by the three-axis acceleration sensor, and the original acceleration signal is subjected to posture compensation and noise elimination by the Kalman compensator to obtain a compensated acceleration data stream, and the gravity component of the compensated acceleration data stream is separated to obtain a pure inertial acceleration sequence; wherein the pure inertial acceleration sequence includes a vertical motion component, a horizontal motion component and a rotational motion component; The pure inertial acceleration sequence is subjected to coordinate system conversion and posture calculation by a preset quaternion transformation technology to obtain a human motion posture parameter set, and a motion trajectory is reconstructed based on the human motion posture parameter set to obtain a three-dimensional space motion trajectory sequence; wherein the three-dimensional space motion trajectory sequence includes a position coordinate sequence, a velocity vector sequence and an acceleration vector sequence; Based on a multi-scale morphological analysis method, feature extraction is performed on the three-dimensional space motion trajectory sequence to obtain a motion morphology feature group, and the motion morphology feature group is transformed into a spectrum domain to obtain a motion spectrum feature sequence; wherein the motion spectrum feature sequence includes a main frequency component, a harmonic component and a modulation component; The motion frequency spectrum feature sequence is subjected to dimension reduction and feature fusion through singular value decomposition to obtain a motion acceleration feature vector.

[0033] Specifically, the inertial parameters of the original acceleration signal during human motion are extracted by the three-axis acceleration sensor to obtain the motion acceleration feature vector. This process first relies on the high-precision three-axis acceleration sensor and its built-in Kalman compensator. In actual operation, the three-axis acceleration sensor is arranged at the key parts of the human body to capture the original acceleration signal during motion in real time. These sensors can record the acceleration changes in three orthogonal directions, namely the acceleration data in the front and back, left and right, and up and down directions. However, due to the changes in external environmental noise and human posture, the original acceleration signal often contains a large amount of noise and interference components. In order to improve the quality of the signal, the Kalman compensator is used to perform posture compensation and noise elimination on these original acceleration signals, so as to obtain the compensated acceleration data stream. The Kalman compensator can effectively suppress noise and correct posture errors by dynamically adjusting the filtering parameters, so that the compensated acceleration data is more accurate and reliable. Next, after obtaining the compensated acceleration data stream, the further step is to separate the gravity component of the data stream to obtain a pure inertial acceleration sequence. Specifically, gravity component separation is to remove the influence of the earth's gravity from the acceleration signal, and only retain the pure inertial acceleration caused by human motion. The purpose of this is to more accurately analyze the acceleration changes of the human body in different motion states. The pure inertial acceleration sequence includes vertical motion components, horizontal motion components, and rotational motion components, which reflect the motion characteristics of the human body in various directions. For example, during running, the vertical motion component can reveal the impact force when jumping and landing, the horizontal motion component shows the speed change of forward and backward, and the rotational motion component describes the rotation of the body. Subsequently, in order to better understand the posture and trajectory of human motion, the pure inertial acceleration sequence is transformed into a coordinate system and the posture is solved through the preset quaternion transformation technology to obtain the human motion posture parameter set, and the motion trajectory is reconstructed based on this parameter set to obtain a three-dimensional space motion trajectory sequence. Quaternion transformation technology is an efficient mathematical tool that can effectively deal with rotation problems in three-dimensional space and avoid the universal joint deadlock phenomenon in the traditional Euler angle representation. Through quaternion transformation, acceleration data can be converted from the sensor coordinate system to the local coordinate system or global coordinate system of the human body, thereby realizing accurate calculation of the human body posture. Based on these posture parameters, the motion trajectory of the human body in three-dimensional space can be further reconstructed, including position coordinate sequence, velocity vector sequence and acceleration vector sequence. This information not only describes the spatial position change of the human body, but also reveals the speed and acceleration characteristics of the motion. Furthermore, in order to extract meaningful features from these rich trajectory data, the three-dimensional space motion trajectory sequence is feature extracted based on the multi-scale morphological analysis method to obtain the motion morphological feature group, and these feature groups are transformed into the spectral domain to obtain the motion spectrum feature sequence.The multi-scale morphological analysis method can effectively extract morphological features in trajectory data, such as key points such as crests and troughs, by operating on structural elements at different scales. Then, these morphological features are converted into frequency domain features through spectral domain transformation (such as Fourier transform or wavelet transform), thereby revealing the changing rules of frequency components during motion. The motion spectrum feature sequence includes main frequency components, harmonic components and modulation components. These features not only reflect the basic frequency characteristics of motion, but also reveal complex dynamic change patterns. Finally, in order to simplify the feature dimension and fuse multiple feature information, the motion spectrum feature sequence is dimensionalized and fused by singular value decomposition to obtain the motion acceleration feature vector. Singular value decomposition is a powerful linear algebra tool that can decompose high-dimensional data matrices into low-rank approximate matrices, thereby effectively reducing feature dimensions and retaining the main information. In this way, the most representative feature vectors can be extracted from complex spectral feature sequences. These feature vectors not only contain rich motion information, but also facilitate subsequent data analysis and pattern recognition tasks. For example, in an athlete training scenario, a coach can analyze these motion acceleration feature vectors to evaluate whether the athlete's technical movements are correct, whether there is a risk of overusing certain muscle groups, or whether the training intensity needs to be adjusted. During marathon training, by analyzing the athlete's motion acceleration feature vectors at different stages, their force and motion efficiency can be accurately judged, so as to formulate a more scientific and reasonable training plan. Similarly, in rehabilitation treatment scenarios, doctors can also use these data to monitor the patient's recovery progress and ensure the effectiveness and safety of rehabilitation training. In this way, the three-axis acceleration sensor and its supporting data processing method not only improve the accuracy of human motion state monitoring, but also provide strong support for the design of personalized training programs and health management. Ultimately, these detailed motion acceleration feature vectors lay a solid foundation for further data analysis and algorithm development, making the motion state detection method based on multimodal sensors more complete and efficient.

[0034] In a specific embodiment, the photoplethysmography sensor is used to monitor the original pulse wave signal during exercise in real time to obtain a cardiovascular physiological characteristic sequence, including: The original pulse wave signal during the movement is collected by a photoelectric volumetric pulse wave sensor, and the morphological features of the original pulse wave signal are extracted to obtain a pulse wave morphological feature set; wherein the pulse wave morphological feature set includes a pulse wave peak value, a double peak interval, and a trough elasticity coefficient; Based on wavelet entropy analysis, the pulse wave morphological feature set is dynamically decomposed in the time-frequency domain to obtain a pulse wave spectrum component array, and the pulse wave spectrum component array is subjected to spectrum peak tracking and phase difference accumulation to obtain a heart rate variability data group; wherein the heart rate variability data group includes low-frequency power density, high-frequency power density ratio and spectrum entropy; Performing a long-term correlation analysis on the heart rate variability data set by using a Hurst index calculation method to obtain a cardiac autonomic nerve regulation parameter set, and dynamically estimating blood pressure based on the cardiac autonomic nerve regulation parameter set to obtain a continuous blood pressure change curve; wherein the continuous blood pressure change curve includes systolic pressure, diastolic pressure and mean arterial pressure; The vascular compliance of the continuous blood pressure change curve is evaluated based on a preset fractional-order derivative model to obtain a peripheral vascular resistance index, and a multivariate coupling analysis is performed on the peripheral vascular resistance index and the heart rate variability data group to obtain a cardiovascular physiological characteristic sequence; wherein the cardiovascular physiological characteristic sequence includes a cardiac load index, a vascular regulation responsiveness, and an autonomic nervous balance coefficient.

[0035] Specifically, the original pulse wave signal during exercise is monitored in real time by a photoelectric plethysmography sensor to obtain a cardiovascular physiological feature sequence. This process first relies on a highly sensitive photoelectric plethysmography sensor and its advanced signal processing technology. In actual operation, the photoelectric plethysmography sensor is placed on a specific part of the human body (such as a finger or earlobe) to capture the light intensity changes caused by blood flow during exercise in real time, thereby obtaining the original pulse wave signal. These signals contain rich information, such as cardiovascular physiological parameters such as heart rate, blood oxygen saturation, and vascular elasticity. In order to extract meaningful cardiovascular physiological features from these complex original signals, a series of advanced signal processing methods are usually required. First, the original pulse wave signal during exercise is collected by a photoelectric plethysmography sensor, and morphological features are extracted from these original pulse wave signals to obtain a pulse wave morphological feature set. Specifically, morphological feature extraction is to identify key feature points such as pulse wave peak, double peak interval, and trough elasticity coefficient by waveform analysis of the pulse wave signal. The pulse wave peak value reflects the maximum pressure generated by the heart during each contraction, the double peak interval reveals the temporal characteristics of the pulse wave propagation, and the trough elastic coefficient describes the elastic state of the blood vessel wall. These features not only provide basic information about cardiac function, but also reflect the health of the vascular system. Next, in order to further analyze the dynamic characteristics of the pulse wave signal, the pulse wave morphological feature set is dynamically decomposed in the time-frequency domain based on wavelet entropy analysis to obtain the pulse wave spectrum component array. Wavelet entropy analysis is a powerful tool that can provide high-resolution signal decomposition results in the time-frequency domain at the same time. Through this analysis, the pulse wave signal can be decomposed into components of multiple different frequency bands, and the energy distribution of each component can be calculated. Then, the pulse wave spectrum component array is subjected to spectral peak tracking and phase difference accumulation to obtain the heart rate variability data set. The heart rate variability data set includes low-frequency power density, high-frequency power density ratio and spectral entropy. These indicators not only reflect the changing laws of the heart rhythm, but also reveal the regulatory ability of the autonomic nervous system. For example, in an athlete training scenario, a coach can evaluate the athlete's physical load and recovery by analyzing these heart rate variability data to ensure that the training plan can achieve the desired effect without causing excessive fatigue or injury risks to the athlete. Furthermore, in order to gain a deeper understanding of the regulatory mechanism of the cardiac autonomic nervous system, the Hurst index calculation method is used to perform a long-term correlation analysis on the heart rate variability data group to obtain a set of cardiac autonomic nervous regulation parameters. The Hurst index is a statistic used to measure the long-term memory characteristics of a time series. By calculating this index, the complexity and stability of the heart rhythm can be revealed. Based on the cardiac autonomic nervous regulation parameter set, dynamic blood pressure estimation can be performed to obtain a continuous blood pressure change curve.The continuous blood pressure change curve includes systolic pressure, diastolic pressure and mean arterial pressure. These indicators not only describe the pressure changes during the heart pumping process, but also reflect the resistance and compliance of the vascular system. For example, during marathon training, by analyzing the continuous blood pressure change curves of athletes at different stages, the response of their cardiovascular system can be accurately judged, so as to formulate a more scientific and reasonable training plan. Finally, in order to evaluate the overall health of the vascular system, the vascular compliance of the continuous blood pressure change curve is evaluated based on the preset fractional derivative model to obtain the peripheral vascular resistance index. The fractional derivative model is an advanced mathematical tool that can better describe the behavior of nonlinear dynamic systems. In this way, the peripheral vascular resistance index can be extracted from the continuous blood pressure change curve, which reflects the responsiveness and regulation ability of blood vessels to external stimuli. Then, the peripheral vascular resistance index and the heart rate variability data group are multivariately coupled to obtain the cardiovascular physiological feature sequence. The cardiovascular physiological feature sequence includes the heart load index, vascular regulation reactivity and autonomic nerve balance coefficient. These features not only comprehensively reflect the overall state of the cardiovascular system, but also provide a solid foundation for subsequent data analysis and pattern recognition tasks. For example, in a rehabilitation treatment scenario, doctors can monitor the patient's recovery progress by analyzing the patient's cardiovascular physiological characteristic sequence during rehabilitation training. If it is found that the patient's peripheral vascular resistance index is significantly reduced within a certain period of time and the ratio of low-frequency power density to high-frequency power density in the heart rate variability data group tends to be normal, it indicates that the patient's vascular system is gradually recovering normal function. Doctors can adjust the treatment plan in time based on this information to ensure the effectiveness and safety of rehabilitation training. In this way, the photoelectric volumetric pulse wave sensor and its supporting data processing method not only improve the accuracy of monitoring the human cardiovascular state, but also provide strong support for the design of personalized training programs and health management. Ultimately, these detailed cardiovascular physiological characteristic sequences lay a solid foundation for further data analysis and algorithm development, making the motion state detection method based on multimodal sensors more complete and efficient.

[0036] In a specific embodiment, the vascular compliance evaluation of the continuous blood pressure change curve based on a preset fractional derivative model to obtain a peripheral vascular resistance index includes: The continuous blood pressure change curve is subjected to non-integer derivative decomposition by using a Caputo fractional differential operator to obtain a vascular wall stress-strain relationship curve, and the vascular wall stress-strain relationship curve is subjected to piecewise polynomial fitting to obtain a vascular elastic modulus distribution function; wherein the vascular elastic modulus distribution function includes a large artery compliance parameter, a microcirculatory resistance coefficient, and a vascular wall viscoelastic factor; Based on the preset Windkessel pulsating blood flow model, the vascular elastic modulus distribution function is coupled with hemodynamic parameters to obtain a vascular pressure-volume response characteristic spectrum, and the vascular pressure-volume response characteristic spectrum is subjected to nonlinear regression analysis to obtain a vascular compliance attenuation curve; wherein the vascular compliance attenuation curve includes an exponential attenuation parameter, a time constant matrix, and an elastic recovery coefficient; The energy distribution of the vascular compliance attenuation curve is extracted in the time-frequency domain by using a wavelet packet entropy value calculation method to obtain a vascular function state set, and the pulse pressure conduction velocity is estimated based on the vascular function state set to obtain an arteriosclerosis index; wherein the arteriosclerosis index includes a pulse pressure amplification factor, a reflection wave enhancement factor, and a pulse conduction delay time; The vascular-cardiac feedback mechanism of the arteriosclerosis index and the heart rate variability data group is modeled by a preset fractional derivative model to obtain a peripheral vascular resistance index; wherein the peripheral vascular resistance index includes a vascular tension regulation coefficient, a sympathetic-parasympathetic balance ratio and a vascular reactivity reserve value.

[0037] Specifically, the continuous blood pressure change curve is evaluated for vascular compliance by a preset fractional derivative model to obtain a peripheral vascular resistance index. This process first relies on the Caputo fractional differential operator and its powerful non-integer derivative decomposition capability. In actual operation, the continuous blood pressure change curve reflects the change of blood pressure over time during the heart pumping process. These data contain rich information, such as systolic pressure, diastolic pressure and mean arterial pressure. In order to extract meaningful vascular compliance characteristics from these complex blood pressure change curves, a series of advanced mathematical tools and algorithms are usually required. First, the continuous blood pressure change curve is decomposed by non-integer derivatives by the Caputo fractional differential operator to obtain the vascular wall stress-strain relationship curve. The Caputo fractional differential operator is an advanced mathematical tool that can better describe the behavior of nonlinear dynamic systems, especially when dealing with time series with memory effects. By performing non-integer derivative decomposition on the blood pressure change curve, the stress-strain relationship of the vascular wall at different time scales can be revealed. Then, the obtained vascular wall stress-strain relationship curve is fitted by piecewise polynomial fitting to obtain the vascular elastic modulus distribution function. This function not only includes the compliance parameters of large arteries, but also covers the microcirculatory resistance coefficient and the vascular wall viscoelastic factor, which together describe the overall elasticity and resistance characteristics of the vascular system. For example, in the athlete training scenario, the coach can evaluate the cardiovascular health of the athlete by analyzing these vascular elastic modulus distribution functions, and formulate a more scientific and reasonable training plan based on this. Next, the vascular elastic modulus distribution function is coupled with the hemodynamic parameters based on the preset Windkessel pulsating blood flow model to obtain the vascular pressure-volume response characteristic spectrum. The Windkessel model is a classic hemodynamic model that describes the dynamic behavior of blood flow by simulating the impedance characteristics of the vascular system. By combining the vascular elastic modulus distribution function with the Windkessel model, the pressure-volume response characteristics of the vascular system can be more comprehensively understood. Specifically, by performing nonlinear regression analysis on the vascular pressure-volume response characteristic spectrum, the vascular compliance attenuation curve can be obtained. The curve includes exponential attenuation parameters, time constant matrix and elastic recovery coefficient. These indicators not only reflect the dynamic change law of vascular compliance, but also reveal the long-term regulatory ability of the vascular system. For example, during marathon training, by analyzing the vascular compliance attenuation curves of athletes at different stages, we can accurately determine the response of their cardiovascular system, thereby formulating a more scientific and reasonable training plan. Furthermore, in order to gain a deeper understanding of the vascular functional state, the wavelet packet entropy calculation method is used to extract the energy distribution of the vascular compliance attenuation curve in the time and frequency domains to obtain the vascular functional state set. The wavelet packet entropy calculation method is an efficient signal processing technology that can provide high-resolution energy distribution results in both the time and frequency domains.Through this method, rich energy distribution information can be extracted from the vascular compliance attenuation curve, which not only describes the instantaneous state of the vascular system, but also reveals its long-term evolution trend. Then, the pulse pressure conduction velocity is estimated based on the vascular function state set to obtain the arteriosclerosis index. The arteriosclerosis index includes the pulse pressure amplification factor, the reflection wave enhancement factor, and the pulse conduction delay time. These indicators not only reflect the degree of arteriosclerosis, but also reveal the overall health of the vascular system. For example, in the rehabilitation treatment scenario, doctors can monitor the patient's recovery progress by analyzing the arteriosclerosis index of the patient during rehabilitation training to ensure the effectiveness and safety of rehabilitation training. Finally, in order to comprehensively evaluate the feedback mechanism between blood vessels and the heart, the vascular-heart feedback mechanism is modeled by the preset fractional derivative model for the arteriosclerosis index and the heart rate variability data set to obtain the peripheral vascular resistance index. This process relies on the powerful modeling ability of the fractional derivative model, which can effectively capture the complex interaction between blood vessels and the heart. Specifically, by combining the arteriosclerosis index with the heart rate variability data set, a complete vascular-heart feedback mechanism model can be established. The model not only includes the vascular tension regulation coefficient, but also covers the sympathetic-parasympathetic balance ratio and vascular reactivity reserve value, which together describe the overall regulatory capacity and stability of the cardiovascular system. For example, in the athlete training scenario, the coach can evaluate the athlete's physical load and recovery by analyzing these peripheral vascular resistance indexes to ensure that the training plan can achieve the expected results without causing excessive fatigue or injury risks to the athlete. To give a specific example, in the long jump event, the athlete's muscle activity, body posture changes, and cardiovascular response at the moment of take-off are all key factors affecting the performance. The peripheral vascular resistance index obtained through the above steps can record the changes in these key parameters in detail. If it is found that the athlete's peripheral vascular resistance index increases significantly within a certain period of time, it means that his vascular system may be in a high state of tension, which may lead to increased heart burden and decreased blood circulation efficiency. The coach can adjust the training intensity or arrange appropriate rest time in time based on this information to avoid potential health risks and optimize sports performance. For example, in the rehabilitation treatment scenario, doctors can also use this method to monitor the patient's recovery progress. By analyzing the patient's peripheral vascular resistance index during rehabilitation training, the recovery of his cardiovascular system can be dynamically monitored. For example, by observing the changes in the peripheral vascular resistance index over a certain period of time, it can be determined whether the patient has the ability to perform higher-intensity training. If it is found that the patient's peripheral vascular resistance index gradually returns to normal over a period of time and the ratio of low-frequency power density to high-frequency power density in the heart rate variability data group tends to be normal, it indicates that the patient's vascular system is gradually recovering its normal function. Doctors can adjust the treatment plan in a timely manner based on this information to ensure the effectiveness and safety of rehabilitation training.In this way, the process of evaluating vascular compliance of continuous blood pressure change curves based on a preset fractional-order derivative model not only improves the accuracy of monitoring the cardiovascular state of the human body, but also provides strong support for the design of personalized training programs and health management. Ultimately, these detailed peripheral vascular resistance indices lay a solid foundation for further data analysis and algorithm development, making the motion state detection method based on multimodal sensors more complete and efficient. In addition, considering individual differences, the specific conditions of different athletes or patients may be different. Therefore, in actual applications, the sensor layout and data analysis methods need to be appropriately adjusted according to individual characteristics to better meet the needs of different application scenarios.

[0038] In a specific embodiment, the adaptive wavelet transform technology is used to perform a time-frequency domain joint analysis on the time-series electromyographic feature data set, the motion acceleration feature vector and the cardiovascular physiological feature sequence to obtain a multi-dimensional feature fusion matrix, including: Adopting adaptive wavelet transform technology to perform time-frequency decomposition on the time-series electromyographic feature data set to obtain an electromyographic signal wavelet coefficient matrix, and performing energy distribution calculation on the electromyographic signal wavelet coefficient matrix to obtain an electromyographic activity energy density spectrum; Performing multi-scale decomposition on the motion acceleration feature vector by using adaptive wavelet transform technology to obtain a motion acceleration wavelet coefficient array, and performing singular spectrum analysis on the motion acceleration wavelet coefficient array to obtain a motion dynamics spectrum map; Based on the adaptive wavelet transform technology, a time-varying spectrum analysis is performed on the cardiovascular physiological characteristic sequence to obtain a cardiovascular response wavelet coefficient group, and a bispectral coherence estimation is performed on the cardiovascular response wavelet coefficient group to obtain a cardiopulmonary coupling characteristic spectrum; The electromyographic activity energy density map, the motion dynamics spectrum map and the cardiopulmonary coupling feature map are aligned in time series by Hilbert-Huang transform to obtain a multimodal feature synchronization matrix, and the multimodal feature synchronization matrix is ​​subjected to high-order tensor decomposition to obtain a multidimensional feature fusion matrix; wherein the multidimensional feature fusion matrix includes motion-cardiovascular coupling mode, electromyographic-acceleration synergy parameters and physiological response integration feature space.

[0039] Specifically, the adaptive wavelet transform technology is used to perform a joint analysis of the time-frequency domain of the time-series electromyographic feature data set, the motion acceleration feature vector and the cardiovascular physiological feature sequence to obtain a multi-dimensional feature fusion matrix. This process first relies on the powerful signal processing capability of the adaptive wavelet transform technology. In actual operation, the adaptive wavelet transform technology is used to decompose and analyze different types of raw data, which can effectively extract the key features in each signal and capture their changing laws on different time scales. First, the adaptive wavelet transform technology is used to perform time-frequency decomposition on the time-series electromyographic feature data set to obtain the electromyographic signal wavelet coefficient matrix, and the energy distribution of the electromyographic signal wavelet coefficient matrix is ​​calculated to obtain the electromyographic activity energy density spectrum. Specifically, the adaptive wavelet transform can automatically adjust its basis function according to the characteristics of the input signal, thereby providing high-resolution analysis results in both the time domain and the frequency domain. By performing time-frequency decomposition on the time-series electromyographic feature data set, a series of wavelet coefficients can be obtained, which contain the information of the signal in different frequency bands. Furthermore, by calculating the energy distribution of these wavelet coefficients, an EMG energy density map can be generated, which not only reflects the basic information of muscle activity, but also reveals the activation mode and conduction velocity changes of muscle fibers under different motion states. For example, in an athlete training scenario, a coach can analyze these EMG energy density maps to evaluate whether the athlete's technical movements are correct, whether there is a risk of overusing certain muscle groups, or whether the training intensity needs to be adjusted to optimize cardiopulmonary function. Next, the motion acceleration feature vector is multi-scale decomposed by adaptive wavelet transform technology to obtain a motion acceleration wavelet coefficient array, and the motion acceleration wavelet coefficient array is subjected to singular spectrum analysis to obtain a motion dynamics spectrum map. Multi-scale decomposition is a method that can reflect the dynamic characteristics of a signal at different scales, while nonlinear singular spectrum analysis can reveal the complex dynamic characteristics hidden inside the signal. By combining these two techniques, rich modulation features such as changes in frequency components and fluctuations in signal envelopes can be extracted from the original motion acceleration feature vector. These motion dynamics spectrum maps not only contain the basic information of the motion, but also reflect the changes in body posture and speed under different motion states. For example, during marathon training, by analyzing the athlete's motion dynamics spectrum at different stages, their force and motion efficiency can be accurately judged, so as to formulate a more scientific and reasonable training plan. Subsequently, based on the adaptive wavelet transform technology, the cardiovascular physiological characteristic sequence is subjected to time-varying spectrum analysis to obtain a cardiovascular response wavelet coefficient group, and the cardiovascular response wavelet coefficient group is subjected to bispectral coherence estimation to obtain a cardiopulmonary coupling characteristic spectrum. Time-varying spectrum analysis is a method that can reflect the frequency characteristics of a signal that changes over time, while bispectral coherence estimation can reveal the nonlinear interactions between signals.By combining these two technologies, rich time-frequency features can be extracted from the cardiovascular physiological feature sequence, which not only describes the instantaneous state of the cardiovascular system, but also reveals its long-term evolution trend. For example, in the rehabilitation treatment scenario, doctors can monitor the patient's recovery progress by analyzing the cardiopulmonary coupling feature map of the patient during rehabilitation training to ensure the effectiveness and safety of rehabilitation training. In order to effectively integrate the above three different types of data (myoelectric activity energy density map, motion dynamics spectrum map and cardiopulmonary coupling feature map), these maps are aligned in time series by Hilbert-Huang transform to obtain a multimodal feature synchronization matrix. Hilbert-Huang transform is an advanced time-frequency analysis tool that can effectively process non-stationary signals and is particularly suitable for the analysis of biomedical signals. By aligning these maps in time series, it can be ensured that data of different modes are aligned on the same time axis, thereby realizing the synchronous analysis of multimodal features. Then, the multimodal feature synchronization matrix is ​​subjected to high-order tensor decomposition to obtain a multi-dimensional feature fusion matrix. High-order tensor decomposition is a powerful mathematical tool that can decompose high-dimensional data into low-rank approximate matrices, thereby effectively reducing feature dimensions and retaining main information. Through this method, the most representative feature vectors can be extracted from the complex multimodal feature synchronization matrix. These feature vectors not only contain rich motion information, but also facilitate subsequent data analysis and pattern recognition tasks. Finally, the multidimensional feature fusion matrix includes motion-cardiovascular coupling mode, electromyography-acceleration synergy parameters and physiological response integration feature space. These features not only comprehensively reflect the overall state of the human body during exercise, but also lay a solid foundation for subsequent data analysis and algorithm development. For example, in the long jump event, the athlete's muscle activity, body posture changes and cardiovascular response at the moment of take-off are all key factors affecting the performance. The multidimensional feature fusion matrix obtained through the above steps can record the changes of these key parameters in detail. If it is found that the athlete's cardiovascular response is abnormal or the muscle fatigue increases rapidly within a certain period of time, it means that the athlete may be close to his limit load. The coach can adjust the training intensity or arrange appropriate rest time in time based on this information. For example, in the rehabilitation treatment scenario, doctors can also use this method to monitor the patient's recovery progress. By analyzing the multidimensional feature fusion matrix of the patient in rehabilitation training, the recovery of his muscle function and cardiovascular system can be dynamically monitored. For example, by observing the changes in the exercise-cardiovascular coupling pattern and the electromyographic-acceleration synergy parameters within a certain period of time, it can be determined whether the patient has the ability to perform higher-intensity training. If it is found that the patient's physiological response integration feature space tends to be normal over a period of time, it indicates that the patient's muscles and cardiovascular system are gradually recovering normal function. Doctors can adjust the treatment plan in a timely manner based on this information to ensure the effectiveness and safety of rehabilitation training.In this way, the multi-dimensional feature fusion method based on adaptive wavelet transform technology and Hilbert-Huang transform not only improves the accuracy of human motion state monitoring, but also provides strong support for the design of personalized training programs and health management. Ultimately, these detailed multi-dimensional feature fusion matrices lay a solid foundation for further data analysis and algorithm development, making the motion state detection method based on multimodal sensors more complete and efficient. In addition, considering individual differences, the specific conditions of different athletes or patients may be different. Therefore, in practical applications, it is also necessary to appropriately adjust the sensor layout and data analysis methods according to individual characteristics to better meet the needs of different application scenarios.

[0040] In a specific embodiment, the multi-dimensional feature fusion matrix is ​​dynamically segmented by a hierarchical clustering algorithm to obtain a motion state feature subspace, including: Performing density peak detection on the multi-dimensional feature fusion matrix through adaptive threshold segmentation to obtain a feature density topological distribution map, and performing kernel density contour extraction on the feature density topological distribution map to obtain a multi-dimensional feature clustering boundary set, wherein the multi-dimensional feature clustering boundary set includes an exercise intensity boundary, a cardiovascular response boundary, and an electromyographic activity boundary; Based on the minimum spanning tree technology, hierarchical distance calculation is performed on the multidimensional feature clustering boundary set to obtain a feature space hierarchical structure, and the feature space hierarchical structure is adaptively truncated to obtain a dynamic hierarchical tree structure, wherein the dynamic hierarchical tree structure includes main motion mode nodes, transition state nodes and abnormal state nodes; The dynamic hierarchical tree structure is subjected to Laplace matrix decomposition by a spectral clustering method to obtain a characteristic subspace orthogonal basis, and the characteristic subspace orthogonal basis is subjected to manifold consistency evaluation to obtain a state classification judgment boundary; wherein the state classification judgment boundary includes a steady-state motion region, an acceleration transition region, and a fatigue deceleration region; Based on a preset kernel principal component analysis method, the state classification discrimination boundary is nonlinearly mapped to obtain a motion state feature projection matrix, and high-order statistics are extracted from the motion state feature projection matrix through a hierarchical clustering algorithm to obtain a motion state feature subspace; wherein the motion state feature subspace includes a static state feature set, a uniform motion feature set, a variable speed motion feature set and a fatigue state feature set.

[0041] Specifically, the multi-dimensional feature fusion matrix is ​​dynamically segmented by a hierarchical clustering algorithm to obtain a motion state feature subspace. This process first relies on a variety of advanced data processing and clustering technologies such as adaptive threshold segmentation, kernel density contour extraction, minimum spanning tree technology, spectral clustering method and kernel principal component analysis. In actual operation, these technologies are applied to the multi-dimensional feature fusion matrix in turn to achieve fine classification and recognition of complex motion states. First, the multi-dimensional feature fusion matrix is ​​subjected to density peak detection by adaptive threshold segmentation to obtain a feature density topological distribution map, and the kernel density contour extraction is performed on the feature density topological distribution map to obtain a multi-dimensional feature clustering boundary set. Specifically, adaptive threshold segmentation is a method that can automatically adjust the threshold according to the local density of the data, thereby effectively detecting the peak points of the feature density. These peak points represent high-density areas in the data, usually corresponding to different motion modes or states. By extracting the kernel density contour of the feature density topological distribution map, the boundaries between different feature clusters can be further clarified. The multidimensional feature clustering boundary set includes the boundary of exercise intensity, the boundary of cardiovascular response and the boundary of electromyographic activity. These boundaries not only reflect the distinguishing features between different exercise states, but also reveal the intrinsic connection between them. For example, in the athlete training scenario, the coach can evaluate whether the athlete's technical movements are correct, whether there is a risk of overusing certain muscle groups, or whether the training intensity needs to be adjusted to optimize cardiopulmonary function by analyzing these multidimensional feature clustering boundary sets. Next, the hierarchical distance calculation of the multidimensional feature clustering boundary set is performed based on the minimum spanning tree technology to obtain the feature space hierarchical structure, and the feature space hierarchical structure is adaptively truncated to obtain a dynamic hierarchical tree structure. The minimum spanning tree technology is an effective method for constructing an acyclic weighted graph. By calculating the shortest path between each feature cluster, a hierarchical distance graph reflecting the relationship between feature clusters can be constructed. By adaptively truncating the graph, the optimal hierarchical structure can be dynamically determined according to the actual situation of the data to form a dynamic hierarchical tree structure. The dynamic hierarchical tree structure includes main movement mode nodes, transition state nodes and abnormal state nodes. These nodes not only describe the basic characteristics of different movement states, but also reveal the conversion paths between them. For example, during marathon training, by analyzing the dynamic hierarchical tree structure of athletes at different stages, their force and exercise efficiency can be accurately judged, so as to formulate a more scientific and reasonable training plan. Furthermore, in order to have a deeper understanding of the relationship between feature clusters, the dynamic hierarchical tree structure is subjected to Laplace matrix decomposition through spectral clustering method to obtain the feature subspace orthogonal basis, and the manifold consistency of the feature subspace orthogonal basis is evaluated to obtain the state classification discrimination boundary.The spectral clustering method is a clustering technology based on graph theory. By constructing a similarity matrix and performing Laplace matrix decomposition on it, complex high-dimensional data can be mapped into a low-dimensional space, so that the intrinsic structure of the data can be more easily discovered. By performing manifold consistency evaluation on the orthogonal basis of the feature subspace, the separation and consistency between different feature clusters can be ensured, so as to obtain an accurate state classification discrimination boundary. The state classification discrimination boundary includes the steady-state motion region, the acceleration transition region, and the fatigue deceleration region. These boundaries not only describe the spatial distribution of different motion states, but also reveal the dynamic change rules between them. For example, in the rehabilitation treatment scenario, doctors can monitor the patient's recovery progress by analyzing the state classification discrimination boundary of the patient in rehabilitation training to ensure the effectiveness and safety of rehabilitation training. Finally, based on the preset kernel principal component analysis method, the state classification discrimination boundary is nonlinearly mapped to obtain the motion state feature projection matrix, and the motion state feature projection matrix is ​​extracted by high-order statistics through the hierarchical clustering algorithm to obtain the motion state feature subspace. Kernel principal component analysis is an advanced data dimensionality reduction technology that can project high-dimensional data into a low-dimensional space through nonlinear mapping, thereby retaining the main information of the data. By performing nonlinear mapping on the state classification discrimination boundary, the motion state feature projection matrix can be obtained, which not only contains rich motion information, but also facilitates subsequent clustering analysis. Then, the motion state feature projection matrix is ​​extracted by a hierarchical clustering algorithm to obtain the motion state feature subspace. The motion state feature subspace includes a static state feature set, a uniform motion feature set, a variable speed motion feature set, and a fatigue state feature set. These features not only comprehensively reflect the overall state of the human body during exercise, but also provide a solid foundation for subsequent data analysis and pattern recognition tasks. For example, in the long jump event, the athlete's muscle activity, body posture changes, and cardiovascular reactions at the moment of take-off are all key factors affecting the performance. The motion state feature subspace obtained by the above steps can record the changes of these key parameters in detail. If it is found that the athlete's fatigue state feature set increases significantly within a certain period of time, it means that the athlete may be close to his limit load. The coach can adjust the training intensity or arrange appropriate rest time in time based on this information to avoid potential health risks and optimize sports performance. For example, in the rehabilitation treatment scenario, doctors can also use this method to monitor the patient's recovery progress. By analyzing the motion state feature subspace of patients during rehabilitation training, the recovery of their muscle function and cardiovascular system can be dynamically monitored. For example, by observing the changes in the static state feature set and the uniform motion feature set within a certain period of time, it can be determined whether the patient has the ability to perform higher intensity training. If it is found that the patient's fatigue state feature set tends to be normal within a certain period of time, it indicates that the patient's muscle and cardiovascular system are gradually recovering normal function.Doctors can adjust treatment plans in a timely manner based on this information to ensure the effectiveness and safety of rehabilitation training. In this way, the motion state feature subspace partitioning method based on hierarchical clustering algorithms and a variety of advanced data analysis techniques not only improves the accuracy of human motion state monitoring, but also provides strong support for the design of personalized training programs and health management. Ultimately, these detailed motion state feature subspaces lay a solid foundation for further data analysis and algorithm development, making the motion state detection method based on multimodal sensors more complete and efficient. In addition, considering individual differences, the specific circumstances of different athletes or patients may be different. Therefore, in practical applications, the sensor layout and data analysis methods need to be appropriately adjusted according to individual characteristics to better meet the needs of different application scenarios.

[0042] The above describes the motion state detection method based on the multimodal sensor in the embodiment of the present invention. The following describes the motion state detection device based on the multimodal sensor in the embodiment of the present invention. Figure 2 In one embodiment of the present invention, a motion state detection device based on a multimodal sensor includes: The acquisition module 21 is used to use a bioelectric sensor array to perform multi-point acquisition of electromyographic signals during human motion to obtain a time-series electromyographic feature data set; An extraction module 22 is used to extract inertial parameters from the original acceleration signal during human motion using a three-axis acceleration sensor to obtain a motion acceleration feature vector; A monitoring module 23, for real-time monitoring of the original pulse wave signal during exercise based on a photoplethysmographic sensor to obtain a cardiovascular physiological characteristic sequence; An analysis module 24 is used to perform a time-frequency domain joint analysis on the time-series electromyographic feature data set, the motion acceleration feature vector and the cardiovascular physiological feature sequence using an adaptive wavelet transform technique to obtain a multi-dimensional feature fusion matrix; The segmentation module 25 is used to dynamically segment the multi-dimensional feature fusion matrix through a hierarchical clustering algorithm to obtain a motion state feature subspace.

[0043] In this embodiment, for the specific implementation of each unit in the above device embodiment, please refer to the above method embodiment, which will not be repeated here.

[0044] Reference Figure 3 The present invention also provides a computer device in an embodiment, wherein the internal structure of the computer device can be as follows: Figure 3As shown. The computer device includes a processor, a memory, a display screen, an input device, a network interface and a database connected through a system bus. Among them, the processor designed by the computer is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store the corresponding data in this embodiment. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, the above method is implemented.

[0045] Those skilled in the art will understand that Figure 3 The structure shown in the figure is merely a block diagram of a portion of the structure related to the solution of the present invention, and does not constitute a limitation on the computer device to which the solution of the present invention is applied.

[0046] An embodiment of the present invention further provides a computer-readable storage medium on which a computer program is stored, and when the computer program is executed by a processor, the above method is implemented. It can be understood that the computer-readable storage medium in this embodiment can be a volatile readable storage medium or a non-volatile readable storage medium.

[0047] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media provided by the present invention and used in the embodiments may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double-speed data rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM.

[0048] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, device, article or method including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, device, article or method. In the absence of further restrictions, an element defined by the sentence "includes a ..." does not exclude the presence of other identical elements in the process, device, article or method including the element.

[0049] The above description is only a preferred embodiment of the present invention, and does not limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made by using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A motion state detection method based on a multimodal sensor, characterized in that: The invention is applied to a motion state detection device, wherein the motion state detection device includes a bioelectric sensor array, a three-axis acceleration sensor and a photoelectric volume pulse wave sensor, and comprises the following steps: The bioelectric sensor array is used to collect the electromyographic signals at multiple points during human motion to obtain a time-series electromyographic feature data set. The inertial parameters of the original acceleration signal during human motion are extracted through a three-axis acceleration sensor to obtain a motion acceleration feature vector; Based on the photoplethysmography sensor, the original pulse wave signal during exercise is monitored in real time to obtain the cardiovascular physiological characteristic sequence; Adopting adaptive wavelet transform technology to perform time-frequency domain joint analysis on the time-series electromyographic feature data set, the motion acceleration feature vector and the cardiovascular physiological feature sequence to obtain a multi-dimensional feature fusion matrix; The multi-dimensional feature fusion matrix is ​​dynamically segmented through a hierarchical clustering algorithm to obtain a motion state feature subspace.

2. The motion state detection method based on multimodal sensor according to claim 1, characterized in that: The bioelectric sensor array is provided with a differential amplifier circuit, and the bioelectric sensor array is used to collect multi-point electromyographic signals during human body movement to obtain a time-series electromyographic characteristic data set, including: The electromyographic signals during human motion are collected by a bioelectric sensor array, and the electromyographic signals are pre-amplified and common-mode suppressed by a differential amplifier circuit to obtain a multi-channel electromyographic signal sequence; Performing instantaneous frequency analysis and nonlinear demodulation on the multi-channel electromyographic signal sequence to obtain an electromyographic signal modulation feature set; The second-order moment analysis method is used to extract time-varying features and calculate spatial correlation of the EMG signal modulation feature set to obtain a time-series EMG feature data set; wherein the time-series EMG feature data set includes motor unit action potential, muscle fiber conduction velocity and muscle fatigue data.

3. The motion state detection method based on multimodal sensor according to claim 1, characterized in that: The method of extracting inertial parameters from the original acceleration signal during human motion by a three-axis acceleration sensor to obtain a motion acceleration feature vector includes: The original acceleration signal during human motion is collected by the three-axis acceleration sensor, and the original acceleration signal is subjected to posture compensation and noise elimination by the Kalman compensator to obtain a compensated acceleration data stream, and the gravity component of the compensated acceleration data stream is separated to obtain a pure inertial acceleration sequence; wherein the pure inertial acceleration sequence includes a vertical motion component, a horizontal motion component and a rotational motion component; The pure inertial acceleration sequence is subjected to coordinate system conversion and posture calculation by a preset quaternion transformation technology to obtain a human motion posture parameter set, and a motion trajectory is reconstructed based on the human motion posture parameter set to obtain a three-dimensional space motion trajectory sequence; wherein the three-dimensional space motion trajectory sequence includes a position coordinate sequence, a velocity vector sequence and an acceleration vector sequence; Based on a multi-scale morphological analysis method, feature extraction is performed on the three-dimensional space motion trajectory sequence to obtain a motion morphology feature group, and the motion morphology feature group is transformed into a spectrum domain to obtain a motion spectrum feature sequence; wherein the motion spectrum feature sequence includes a main frequency component, a harmonic component and a modulation component; The motion frequency spectrum feature sequence is subjected to dimension reduction and feature fusion through singular value decomposition to obtain a motion acceleration feature vector.

4. The motion state detection method based on multimodal sensor according to claim 1, characterized in that: The photoplethysmography sensor is used to monitor the original pulse wave signal during exercise in real time to obtain a cardiovascular physiological characteristic sequence, including: The original pulse wave signal during the movement is collected by a photoelectric volumetric pulse wave sensor, and the morphological features of the original pulse wave signal are extracted to obtain a pulse wave morphological feature set; wherein the pulse wave morphological feature set includes a pulse wave peak value, a double peak interval, and a trough elasticity coefficient; Based on wavelet entropy analysis, the pulse wave morphological feature set is dynamically decomposed in the time-frequency domain to obtain a pulse wave spectrum component array, and the pulse wave spectrum component array is subjected to spectrum peak tracking and phase difference accumulation to obtain a heart rate variability data group; wherein the heart rate variability data group includes low-frequency power density, high-frequency power density ratio and spectrum entropy; Performing a long-term correlation analysis on the heart rate variability data set by using a Hurst index calculation method to obtain a cardiac autonomic nerve regulation parameter set, and dynamically estimating blood pressure based on the cardiac autonomic nerve regulation parameter set to obtain a continuous blood pressure change curve; wherein the continuous blood pressure change curve includes systolic pressure, diastolic pressure and mean arterial pressure; The vascular compliance of the continuous blood pressure change curve is evaluated based on a preset fractional-order derivative model to obtain a peripheral vascular resistance index, and a multivariate coupling analysis is performed on the peripheral vascular resistance index and the heart rate variability data group to obtain a cardiovascular physiological characteristic sequence; wherein the cardiovascular physiological characteristic sequence includes a cardiac load index, a vascular regulation responsiveness, and an autonomic nervous balance coefficient.

5. The motion state detection method based on multimodal sensor according to claim 4 is characterized in that: The vascular compliance evaluation of the continuous blood pressure change curve based on the preset fractional derivative model to obtain the peripheral vascular resistance index includes: The continuous blood pressure change curve is subjected to non-integer derivative decomposition by using a Caputo fractional differential operator to obtain a vascular wall stress-strain relationship curve, and the vascular wall stress-strain relationship curve is subjected to piecewise polynomial fitting to obtain a vascular elastic modulus distribution function; wherein the vascular elastic modulus distribution function includes a large artery compliance parameter, a microcirculatory resistance coefficient, and a vascular wall viscoelastic factor; Based on the preset Windkessel pulsating blood flow model, the vascular elastic modulus distribution function is coupled with hemodynamic parameters to obtain a vascular pressure-volume response characteristic spectrum, and the vascular pressure-volume response characteristic spectrum is subjected to nonlinear regression analysis to obtain a vascular compliance attenuation curve; wherein the vascular compliance attenuation curve includes an exponential attenuation parameter, a time constant matrix, and an elastic recovery coefficient; The energy distribution of the vascular compliance attenuation curve is extracted in the time-frequency domain by using a wavelet packet entropy value calculation method to obtain a vascular function state set, and the pulse pressure conduction velocity is estimated based on the vascular function state set to obtain an arteriosclerosis index; wherein the arteriosclerosis index includes a pulse pressure amplification factor, a reflection wave enhancement factor, and a pulse conduction delay time; The vascular-cardiac feedback mechanism of the arteriosclerosis index and the heart rate variability data group is modeled by a preset fractional derivative model to obtain a peripheral vascular resistance index; wherein the peripheral vascular resistance index includes a vascular tension regulation coefficient, a sympathetic-parasympathetic balance ratio and a vascular reactivity reserve value.

6. The motion state detection method based on multimodal sensor according to claim 1, characterized in that: The adaptive wavelet transform technology is used to perform a time-frequency domain joint analysis on the time-series electromyographic feature data set, the motion acceleration feature vector and the cardiovascular physiological feature sequence to obtain a multi-dimensional feature fusion matrix, including: Adopting adaptive wavelet transform technology to perform time-frequency decomposition on the time-series electromyographic feature data set to obtain an electromyographic signal wavelet coefficient matrix, and performing energy distribution calculation on the electromyographic signal wavelet coefficient matrix to obtain an electromyographic activity energy density spectrum; Performing multi-scale decomposition on the motion acceleration feature vector by using adaptive wavelet transform technology to obtain a motion acceleration wavelet coefficient array, and performing singular spectrum analysis on the motion acceleration wavelet coefficient array to obtain a motion dynamics spectrum map; Based on the adaptive wavelet transform technology, a time-varying spectrum analysis is performed on the cardiovascular physiological characteristic sequence to obtain a cardiovascular response wavelet coefficient group, and a bispectral coherence estimation is performed on the cardiovascular response wavelet coefficient group to obtain a cardiopulmonary coupling characteristic spectrum; The electromyographic activity energy density map, the motion dynamics spectrum map and the cardiopulmonary coupling feature map are aligned in time series by Hilbert-Huang transform to obtain a multimodal feature synchronization matrix, and the multimodal feature synchronization matrix is ​​subjected to high-order tensor decomposition to obtain a multidimensional feature fusion matrix; wherein the multidimensional feature fusion matrix includes motion-cardiovascular coupling mode, electromyographic-acceleration synergy parameters and physiological response integration feature space.

7. The motion state detection method based on multimodal sensor according to claim 1, characterized in that: The multi-dimensional feature fusion matrix is ​​dynamically segmented by a hierarchical clustering algorithm to obtain a motion state feature subspace, including: Performing density peak detection on the multi-dimensional feature fusion matrix through adaptive threshold segmentation to obtain a feature density topological distribution map, and performing kernel density contour extraction on the feature density topological distribution map to obtain a multi-dimensional feature clustering boundary set, wherein the multi-dimensional feature clustering boundary set includes an exercise intensity boundary, a cardiovascular response boundary, and an electromyographic activity boundary; Based on the minimum spanning tree technology, hierarchical distance calculation is performed on the multidimensional feature clustering boundary set to obtain a feature space hierarchical structure, and the feature space hierarchical structure is adaptively truncated to obtain a dynamic hierarchical tree structure, wherein the dynamic hierarchical tree structure includes main motion mode nodes, transition state nodes and abnormal state nodes; The dynamic hierarchical tree structure is subjected to Laplace matrix decomposition by a spectral clustering method to obtain a characteristic subspace orthogonal basis, and the characteristic subspace orthogonal basis is subjected to manifold consistency evaluation to obtain a state classification judgment boundary; wherein the state classification judgment boundary includes a steady-state motion region, an acceleration transition region, and a fatigue deceleration region; Based on a preset kernel principal component analysis method, the state classification discrimination boundary is nonlinearly mapped to obtain a motion state feature projection matrix, and high-order statistics are extracted from the motion state feature projection matrix through a hierarchical clustering algorithm to obtain a motion state feature subspace; wherein the motion state feature subspace includes a static state feature set, a uniform motion feature set, a variable speed motion feature set and a fatigue state feature set.

8. A motion state detection device based on a multimodal sensor, characterized in that: Applied to a motion state detection device, the motion state detection device includes a bioelectric sensor array, a three-axis acceleration sensor and a photoelectric volume pulse wave sensor, including: The acquisition module is used to use a bioelectric sensor array to perform multi-point acquisition of electromyographic signals during human motion to obtain a time-series electromyographic feature data set; An extraction module is used to extract inertial parameters from the original acceleration signal during human motion using a three-axis acceleration sensor to obtain a motion acceleration feature vector; A monitoring module, which is used to monitor the original pulse wave signal during exercise in real time based on a photoplethysmographic sensor to obtain a cardiovascular physiological characteristic sequence; An analysis module, for performing a time-frequency domain joint analysis on the time-series electromyographic feature data set, the motion acceleration feature vector and the cardiovascular physiological feature sequence using an adaptive wavelet transform technique to obtain a multi-dimensional feature fusion matrix; The segmentation module is used to dynamically segment the multi-dimensional feature fusion matrix through a hierarchical clustering algorithm to obtain a motion state feature subspace.

9. A computer device comprising a memory and a processor, wherein a computer program is stored in the memory, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images