A motion state monitoring method and intelligent sportswear

Abstract

This invention relates to the technical field of smart wearable devices and sports health monitoring, and more specifically, to a method for monitoring exercise status and a smart sports bra. The method includes: synchronously acquiring a user's physiological and motion signals through a multimodal sensor array; preprocessing the acquired physiological and motion signals; removing motion artifacts from each physiological signal using a deep learning model based on a cross-attention mechanism; extracting multi-dimensional physiological and motion features from the denoised physiological and motion signals; fusing and analyzing the extracted physiological and motion features based on a pre-trained artificial intelligence model to assess the user's exercise status in real time; and generating and outputting feedback information based on the assessment results. This invention achieves multi-dimensional synchronous monitoring of exercise physiological status, improves signal quality and assessment accuracy, and can be widely used in sports training, health monitoring, and rehabilitation management.

Description

Technical Field

[0001] This invention relates to the technical field of smart wearable devices and sports health monitoring, and more specifically, to a method for monitoring exercise status and a smart sports underwear system. Background Technology

[0002] With the increasing awareness of fitness among the general public and the growing demand for scientific professional sports training, smart wearable devices are being used more and more widely in the field of sports and health monitoring. Their core requirement is to provide users with sports status assessment, safety warnings and performance optimization suggestions through real-time and accurate collection of physiological and sports data.

[0003] Traditional sports monitoring devices often rely on a single type of sensor (such as an accelerometer or heart rate sensor) for data collection. For example, smart bracelets typically only count steps using acceleration data or monitor heart rate using photoelectric heart rate sensors. They cannot simultaneously capture the physiological responses of multiple systems, such as the cardiovascular, respiratory, and fluid balance systems, and therefore cannot comprehensively reflect the synergistic relationship between physiological load and exercise posture during exercise, leading to biased assessment results.

[0004] Existing technologies attempt to integrate data from multiple sensors. For example, smart sports bras collect basic data through heart rate belts, thermometers, hygrometers, and inertial measurement units (IMUs). However, the types of sensors are limited, and they mainly focus on heart rate, temperature, and movement trajectory, lacking monitoring of deeper physiological parameters such as respiration, lung sounds, and tissue condition.

[0005] Multimodal data is inherently heterogeneous, with significant differences in time-frequency characteristics between different sensors. For example, ECG signals are low-frequency bioelectrical signals, while temperature and humidity sensors and inertial measurement unit (IMU) data are high-frequency motion mechanical signals. Traditional filtering algorithms, such as mean filtering and Kalman filtering, can only suppress noise for a single type of signal and cannot effectively eliminate motion artifacts that interfere with physiological signals. For instance, during strenuous activities such as high-speed running and jumping, body swaying can cause severe distortion in ECG and PPG signals. Traditional algorithms struggle to distinguish between the physiological signals themselves and motion interference components, leading to a significant increase in calculation errors for key parameters such as heart rate and blood oxygen, and a significant decrease in the signal-to-noise ratio.

[0006] Because the data source is singular and the analysis model is simple, the feedback information from existing technologies is general and lacks practicality. It is difficult to distinguish in detail the specific factors affecting athletic performance, and therefore cannot provide targeted improvement suggestions. The accuracy of early warnings is low and the false alarm rate is high, making it difficult to effectively ensure safety during high-intensity sports. Summary of the Invention

[0007] The purpose of this invention is to overcome the shortcomings of existing technologies and provide a method for monitoring motion status and a smart sports bra. Based on a multimodal sensor array, motion artifact removal and multimodal feature fusion are performed to achieve accurate assessment and personalized feedback of motion status. This enables full-dimensional synchronous monitoring of the physiological state of exercise, improves signal quality and assessment accuracy, and can be widely used in sports training, health monitoring and rehabilitation management.

[0008] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is: a motion state monitoring method, comprising the following steps: S1: Synchronously acquire the user's physiological and motion signals through a multimodal sensor array; S2: Preprocess the collected physiological and motion signals; S3: A deep learning model based on cross-attention mechanism to remove motion artifacts from various physiological signals; S4: Extract multi-dimensional physiological and motion features from the denoised physiological and motion signals; S5: Based on a pre-trained large-scale artificial intelligence model, it fuses and analyzes the extracted physiological and motion features to assess the user's motion status in real time. S6: Generate and output feedback information based on the evaluation results.

[0009] The motion state monitoring method of this invention achieves simultaneous capture of multi-dimensional physiological parameters through multi-modal sensor array fusion, providing a data foundation for comprehensive assessment of motion state. It utilizes a deep learning model based on a cross-attention mechanism to extract and remove motion artifacts, dynamically learning the complex nonlinear relationships between motion signals and various physiological signals, effectively separating and removing noise, and significantly improving the signal-to-noise ratio and feature extraction accuracy in high-speed motion environments. Furthermore, it employs a pre-trained large-scale artificial intelligence model to analyze factors affecting current motion performance in real time and provide targeted suggestions, achieving an intelligent leap from "monitoring" to "guidance." This invention realizes multi-dimensional synchronous monitoring of motion physiological state, improves signal quality and assessment accuracy, and can be widely used in sports training, health monitoring, and rehabilitation management.

[0010] Preferably, the physiological signals include signals acquired by an electrocardiogram sensor, a photoplethysmography (PPG) sensor, a lung sound sensor, an impedance sensor, a temperature sensor, and a respiratory wave sensor; the motion signals include signals acquired by an inertial measurement unit.

[0011] Preferably, in step S2, the preprocessing includes: Bandpass filtering is performed on the physiological signals collected by the ECG sensor, photoplethysmography pulse wave sensor, respiratory wave sensor, and lung sound sensor to retain the effective frequency components; Mean filtering is applied to the physiological signals collected by the impedance sensor and temperature sensor to suppress random noise; The gravity component and attitude drift of the motion signal acquired by the inertial measurement unit are compensated by Kalman filtering.

[0012] Preferably, step S3, the step of removing motion artifacts, includes: S301: Divide all signals into equal-time sliding window segments to obtain the corresponding signal segments. ; S302: Input the signal segment into the motion artifact extraction model, and extract the most relevant components to the motion signal within the signal segment, i.e., motion artifacts, through the cross-attention mechanism; S303: Subtract the corresponding motion artifact from the signal segment to obtain the denoised signal.

[0013] Preferably, the calculation method for motion artifact removal includes: ; ; ; ; ; ; In the formula, This indicates the signal segment of the inertial measurement unit; Indicates from The motion state features extracted from them are used as the query matrix; and These respectively represent the signal segments The waveform features extracted are used as the key matrix and value matrix; express and The similarity between them; This indicates that the extraction was performed using the cross-attention mechanism. The features of motion artifacts are used as attention weight matrices; Indicates motion artifacts; Indicates the denoised signal; , , and These are specially trained deep networks.

[0014] Preferably, in step S4, the extracted multi-dimensional physiological features include: heart rate, heart rate variability; blood oxygen saturation; frequency energy percentage of abnormal respiratory sounds, including wheezing and crackling sounds; tissue water content; body temperature; and respiratory rate. The extracted motion features include: step frequency and motion power.

[0015] Preferably, in step S5, the real-time assessment of the user's exercise status includes: assessing exercise performance through cadence and exercise power, and planning exercise plans in real time; assessing cardiovascular system load and cardiac risk through heart rate and heart rate variability; assessing respiratory system load through blood oxygen, respiratory rate, and the proportion of abnormal breath sounds; and assessing the risk of hypothermia or heatstroke through tissue water content and body temperature.

[0016] Preferably, in step S6, the output feedback information includes: data being transmitted to the mobile phone in real time via Bluetooth, and the comprehensive score, risk warning, and historical trend being displayed on the mobile phone interface; and voice warnings and vibration alerts being set.

[0017] The present invention also provides a smart sports bra for performing the above-described motion state monitoring method; comprising: The underwear itself; A multimodal sensor array is embedded in the elastic fabric layer of the underwear body to synchronously collect the user's physiological signals and motion data. The multimodal sensor array includes at least an electrocardiogram sensor, a photoplethysmography pulse wave sensor, a lung sound sensor, an electrical impedance sensor, an inertial measurement unit, a temperature sensor, and a respiratory wave sensor. The data processing module, connected to the multimodal sensor array, is used to preprocess the collected physiological signals and motion data, and remove motion artifacts from each physiological signal through a cross-attention mechanism based on a deep learning model. The feature extraction module is used to extract multi-dimensional physiological and motion features from physiological signals and motion data after data processing. The analysis and feedback module uses a pre-trained artificial intelligence model to fuse and analyze extracted physiological and motion features, assess the user's motion status in real time, and generate corresponding feedback information.

[0018] This invention relates to a smart sports bra that uses a multimodal sensor array to collect signals from electrocardiograms, photoplethysmography (PPG), lung sounds, electrical impedance, inertial measurement units (IMU), temperature, and respiratory waves. This enables the simultaneous capture of multi-dimensional physiological parameters related to cardiovascular, respiratory, metabolic, and biomechanical aspects, providing a data foundation for comprehensive assessment of exercise status. The data processing module preprocesses the multi-source signals and utilizes a deep learning model based on cross-attention to extract and remove motion artifacts. This allows for dynamic learning of the complex nonlinear relationships between motion signals and various physiological signals, effectively separating and removing noise, and significantly improving the signal-to-noise ratio and feature extraction accuracy in high-speed exercise environments. A pre-trained artificial intelligence model analyzes factors influencing current exercise performance in real time and provides targeted suggestions, achieving an intelligent leap from "monitoring" to "guidance." This invention achieves multi-dimensional synchronous monitoring of exercise physiological status, improves signal quality and assessment accuracy, and can be widely used in sports training, health monitoring, and rehabilitation management.

[0019] The electrocardiogram (ECG) sensor is a flexible fabric electrode used to collect heart rate variability and arrhythmia characteristics; the photoplethysmography (PPG) sensor is positioned on the underwear body corresponding to the sternum to monitor blood oxygen saturation; the lung sound sensor is a piezoelectric film attached to the underwear body corresponding to the chest area to capture respiratory sound characteristics; the impedance sensor measures tissue water content through low-frequency current; the inertial measurement unit includes a triaxial accelerometer and a gyroscope to track movement posture, stride frequency, and impact force; the temperature sensor is positioned on the underwear body corresponding to the armpit area to monitor the difference between core body temperature and body surface temperature; the respiratory wave sensor is a strain-sensing fiber woven into the underwear body corresponding to the underbust area to extract respiratory frequency and depth.

[0020] Compared with the prior art, the beneficial effects of this invention are as follows: (1) By collecting electrocardiogram, photoplethysmography pulse wave, lung sounds, electrical impedance, inertial measurement, temperature and respiratory wave signals through a multimodal sensor array, the full-dimensional synchronous capture of exercise physiological states such as cardiovascular, respiratory, metabolic and biomechanical states is realized, providing a data basis for comprehensive assessment of exercise state; (2) The data processing module preprocesses the multi-source signals and uses a deep learning model based on cross-attention mechanism to extract and remove motion artifacts. It can dynamically learn the complex nonlinear relationship between motion signals and various physiological signals, effectively separate and remove noise, and significantly improve the signal-to-noise ratio and feature extraction accuracy in high-speed motion environments. (3) Using pre-trained artificial intelligence large models to analyze factors affecting current sports performance in real time and provide targeted suggestions, realize the intelligent leap from "monitoring" to "guidance", improve the accuracy of assessment, and can be widely used in sports training, health monitoring and rehabilitation management. Attached Figure Description

[0021] Figure 1 This is a flowchart of the motion state monitoring method in an embodiment of the present invention; Figure 2 This is a flowchart illustrating the removal of motion artifacts in an embodiment of the present invention; Figure 3 This is a schematic diagram of the structure of the smart sports bra from a first-view perspective in an embodiment of the present invention; Figure 4 This is a structural schematic diagram of the smart sports bra from a second perspective in an embodiment of the present invention.

[0022] In the attached diagram: 1-Underwear body; 2-ECG sensor; 3-Photoplethysmography (PPG) sensor; 4-Lung sound sensor; 5-Electrical impedance sensor; 6-Inertial measurement unit; 7-Temperature sensor; 8-Respiratory wave sensor; 9-Main control module. Detailed Implementation

[0023] The present invention will be further described below with reference to specific embodiments. The accompanying drawings are for illustrative purposes only, representing schematic diagrams rather than actual physical objects, and should not be construed as limiting the scope of this patent. To better illustrate the embodiments of the present invention, some components in the drawings may be omitted, enlarged, or reduced, and do not represent the actual dimensions of the product. It is understandable to those skilled in the art that some well-known structures and their descriptions may be omitted in the drawings.

[0024] In the accompanying drawings of the embodiments of the present invention, the same or similar reference numerals correspond to the same or similar components. In the description of the present invention, it should be understood that if terms such as "upper," "lower," "left," "right," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, they are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, the terms used to describe positional relationships in the drawings are only for illustrative purposes and should not be construed as limiting the present patent. For those skilled in the art, the specific meaning of the above terms can be understood according to the specific circumstances. Example 1

[0025] A motion state monitoring method, such as Figure 1 As shown, it includes the following steps: S1: Synchronously acquire the user's physiological and motion signals through a multimodal sensor array; S2: Preprocess the collected physiological and motion signals; S3: A deep learning model based on cross-attention mechanism to remove motion artifacts from various physiological signals; S4: Extract multi-dimensional physiological and motion features from the denoised physiological and motion signals; S5: Based on a pre-trained large-scale artificial intelligence model, it fuses and analyzes the extracted physiological and motion features to assess the user's motion status in real time. S6: Generate and output feedback information based on the evaluation results.

[0026] The aforementioned motion state monitoring method, through multimodal sensor array fusion, achieves simultaneous capture of multi-dimensional physiological parameters, providing a data foundation for comprehensive assessment of motion state. Utilizing a deep learning model based on a cross-attention mechanism to extract and remove motion artifacts, it dynamically learns the complex nonlinear relationships between motion signals and various physiological signals, effectively separating and removing noise, and significantly improving the signal-to-noise ratio and feature extraction accuracy in high-speed motion environments. Furthermore, a pre-trained large-scale artificial intelligence model analyzes factors affecting current motion performance in real time and provides targeted suggestions, achieving an intelligent leap from "monitoring" to "guidance." This embodiment achieves multi-dimensional synchronous monitoring of motion physiological state, improving signal quality and assessment accuracy, and can be widely used in sports training, health monitoring, and rehabilitation management.

[0027] Physiological signals include signals acquired by the electrocardiogram sensor 2 (ECG), photoplethysmography (PPG) sensor 3, lung sound sensor 4, electrical impedance sensor 5, temperature sensor 7, and respiratory wave sensor 8; motion signals include signals acquired by the inertial measurement unit 6 (IMU).

[0028] In step S2, the preprocessing includes: The physiological signals collected by ECG sensor 2, photoplethysmography pulse wave sensor 3, respiratory wave sensor 8, and lung sound sensor 4 are bandpass filtered to retain effective frequency components. Mean filtering is applied to the physiological signals collected by impedance sensor 5 and temperature sensor 7 to suppress random noise; The gravity component and attitude drift of the motion signal acquired by the inertial measurement unit 6 are compensated by Kalman filtering.

[0029] In step S3, such as Figure 2 As shown, the steps for removing motion artifacts include: S301: Divide all signals into equal-time sliding window segments to obtain the corresponding signal segments. ; S302: Input the signal segment into the motion artifact extraction model, and extract the most relevant components to the motion signal within the signal segment, i.e., motion artifacts, through the cross-attention mechanism; ; ; ; ; ; This indicates the signal segment of the inertial measurement unit; Indicates from The motion state features extracted from them are used as the query matrix; and These respectively represent the signal segments The waveform features extracted are used as the key matrix and value matrix; express and The similarity between them; This indicates that the extraction was performed using the cross-attention mechanism. The features of motion artifacts are used as attention weight matrices; Indicates motion artifacts; , , and These are specially trained deep networks; S303: Subtract the corresponding motion artifacts from the signal segment to obtain the denoised signal: ; In the formula, This represents the denoised signal.

[0030] In this embodiment, the motion signal of the inertial measurement unit 6 is used as a guide, and the motion-related noise components in each physiological signal are extracted and subtracted through the attention mechanism to improve the signal-to-noise ratio under high-speed motion.

[0031] In step S4, the extracted multidimensional physiological features include: heart rate, heart rate variability; blood oxygen saturation; frequency energy percentage of abnormal respiratory sounds, including wheezing and crackles; tissue water content; body temperature; and respiratory rate. The extracted motion features include: step frequency and motion power.

[0032] In step S5, the real-time assessment of the user's exercise status includes: assessing exercise performance through cadence and exercise power, and planning exercise plans in real time; assessing cardiovascular system load and cardiac risk through heart rate and heart rate variability; assessing respiratory system load through blood oxygen, respiratory rate, and the proportion of abnormal breath sounds; and assessing the risk of hypothermia or heatstroke through tissue water content and body temperature.

[0033] In step S6, the output feedback information includes: data is transmitted to the user terminal, including mobile phone, in real time via Bluetooth, and the comprehensive score, risk warning and historical trend are displayed on the user terminal interface; voice warning and vibration prompt can also be set to ensure unimpeded interaction with peripheral devices during exercise. Example 2

[0034] A smart sports bra for performing the motion state monitoring method of Embodiment 1; such as Figure 3 , Figure 4 As shown, it includes: Underwear body 1; A multimodal sensor array is embedded in the elastic fabric layer of the underwear body 1 to synchronously collect the user's physiological signals and motion data. The multimodal sensor array includes at least an electrocardiogram sensor 2, a photoplethysmography pulse wave sensor 3, a lung sound sensor 4, an electrical impedance sensor 5, an inertial measurement unit 6, a temperature sensor 7, and a respiratory wave sensor 8. The data processing module, connected to the multimodal sensor array, is used to preprocess the collected physiological signals and motion data, and remove motion artifacts from each physiological signal through a cross-attention mechanism based on a deep learning model. The feature extraction module is used to extract multi-dimensional physiological and motion features from physiological signals and motion data after data processing. The analysis and feedback module uses a pre-trained artificial intelligence model to fuse and analyze extracted physiological and motion features, assess the user's motion status in real time, and generate corresponding feedback information.

[0035] The aforementioned smart sports bra collects ECG, photoplethysmography (PPG), lung sounds, electrical impedance, inertial measurement, temperature, and respiratory wave signals through a multimodal sensor array. This enables the simultaneous capture of multi-dimensional physiological parameters, including cardiovascular, respiratory, metabolic, and biomechanical parameters, providing a data foundation for comprehensive assessment of exercise status. The data processing module preprocesses the multi-source signals and utilizes a deep learning model based on cross-attention to extract and remove motion artifacts. This allows for dynamic learning of the complex nonlinear relationships between motion signals and various physiological signals, effectively separating and removing noise, and significantly improving the signal-to-noise ratio and feature extraction accuracy in high-speed exercise environments. A pre-trained artificial intelligence model analyzes factors influencing current exercise performance in real time and provides targeted suggestions, achieving an intelligent leap from "monitoring" to "guidance." This embodiment achieves multi-dimensional synchronous monitoring of exercise physiological status, improving signal quality and assessment accuracy, and can be widely used in sports training, health monitoring, and rehabilitation management.

[0036] like Figure 3 , Figure 4As shown, ECG sensor 2 is a flexible fabric electrode used to collect heart rate variability and arrhythmia characteristics; photoplethysmography (PPG) sensor 3 is positioned on the underwear body 1 corresponding to the sternum to monitor blood oxygen saturation; lung sound sensor 4 is a piezoelectric film attached to the front of the underwear body 1 to capture respiratory sound characteristics; impedance sensor 5 measures tissue water content using low-frequency current; inertial measurement unit 6 includes a triaxial accelerometer and a gyroscope to track movement posture, step frequency, and impact force; temperature sensor 7 is positioned on the underwear body 1 corresponding to the armpit to monitor the difference between core body temperature and body surface temperature; respiratory wave sensor 8 is a strain sensing fiber woven into the underwear body 1 corresponding to the lower chest area to extract respiratory rate and depth. The arrangement of the multimodal sensor array has been optimized to ensure signal quality and wearing comfort.

[0037] It also includes a communication module and a user interaction module. The communication module is used to transmit the processed data and feedback information to the user terminal. The user interaction module includes a display interface, a voice broadcast unit and / or a vibration prompt unit, which are used to output feedback information to the user to ensure unimpeded interaction of peripheral devices during movement.

[0038] In this embodiment, the data processing module, feature extraction module, analysis feedback module, communication module, and user interaction module are integrated into the main control module 9, such as... Figure 3 As shown. Example 3

[0039] This embodiment demonstrates the application of the smart sports underwear in Embodiment Two. A marathon runner wore the smart sports underwear from Embodiment Two during a marathon (approximately 42.195 kilometers). The race environment was outdoor, with moderate temperatures (approximately 15-20 degrees Celsius). However, athletes face risks such as cardiovascular load, decreased respiratory efficiency, dehydration, and elevated body temperature during prolonged high-intensity exercise.

[0040] The multimodal sensor array synchronously acquires data at a set sampling rate: ECG sensor 2 acquires ECG signals at 100Hz to calculate heart rate and heart rate variability (HRV); photoplethysmography (PPG) sensor 3 monitors blood oxygen saturation at 100Hz. The sensor detects respiratory sounds and microcirculation status; lung sound sensor 4 captures respiratory sounds, such as normal respiratory sounds or abnormal sounds (such as wheezing), at 4 kHz; impedance sensor 5 measures tissue water content at 100 Hz via low-frequency current to assess dehydration status; inertial measurement unit (IMU) 6 tracks motion posture, cadence, acceleration, and impact force at 200 Hz; temperature sensor 7 monitors core body temperature and surface temperature difference at 100 Hz; and respiratory wave sensor 8 extracts respiratory frequency and depth at 100 Hz.

[0041] The raw signals are preprocessed as follows: ECG and PPG signals are bandpass filtered (0.5-4.0Hz) to retain the effective heart rate component; respiratory waves are bandpass filtered (0.1-20Hz) to highlight respiratory characteristics; lung sound signals are bandpass filtered (10-2kHz) to retain the ability to capture high-frequency noise; IMU data are compensated for gravity components and attitude drift by Kalman filtering to improve the accuracy of motion trajectory; impedance and temperature signals are mean filtered (window size 1 second) to reduce random noise.

[0042] Motion artifact removal: All signals are segmented into sliding windows at equal time intervals (window length 4s, corresponding to the sampling rate) to obtain signal segments. , This includes EPC, PPG, lung sounds, electrical impedance and temperature, respiratory waves, and IMU; for each signal segment, a deep learning model based on a cross-attention mechanism is invoked, with the model using the IMU data segment of the current time window. As input, through embedded networks Its motion features are extracted as the query matrix Q, and simultaneously embedded through a network. , Extract waveform features of the signal segment to be processed as a key matrix Sum matrix Calculate the attention weight matrix And by reconstructing the network Generate estimated motion artifacts Finally, the corresponding motion artifacts are subtracted from the signal segment to obtain the denoised signal.

[0043] Key features were extracted from each denoised signal: heart rate and heart rate variability; blood oxygen saturation; frequency energy percentage of abnormal breathing sounds, including wheezing and crackling sounds; tissue water content; body temperature; respiratory rate; cadence and exercise power.

[0044] Multimodal data analysis and feedback are performed using large AI models (such as a finely tuned Qwen 7B model): By combining IMU cadence and power data with historical trends, the athlete's current pace can be assessed in real time. For example, in the middle of the race (about 20 kilometers), if the cadence drops from 180 steps / minute to 170 steps / minute and the power decreases, indicating fatigue accumulation, the user terminal interface will display "Pace has decreased, it is recommended to adjust to the target pace of 5:30 / km" and provide voice prompts to the athlete. Combining ECG and PPG data, the heart rate consistently exceeded 160 bpm while HRV decreased. When the heart rate drops to 92%, it is judged to be a potential risk of hypoxia. The user terminal displays "Heart load is too high. It is recommended to slow down and take deep breaths," triggering a vibration warning. Lung sound sensor 4 detected wheezing (frequency energy percentage exceeds threshold) and the respiratory rate rose to 28 breaths / min, indicating low respiratory efficiency. The voice prompt said, "Rapid breathing detected, it is recommended to adjust the breathing rhythm to 2-step inhale / 2-step exhale." Electrical impedance analysis showed that tissue water content decreased from 70% to 65%, and body temperature rose to 38.5°C. Based on the ambient temperature, the system determined that there was a risk of dehydration and heatstroke. The user terminal displayed "High risk of dehydration, it is recommended to replenish electrolyte drinks" and provided information on nearby water replenishment points. The user terminal interface displays the athlete's status in the form of a comprehensive score (0-100 points) and generates a historical trend chart. At critical moments (such as at 30 kilometers), the system comprehensively predicts the risk of "hitting the wall" based on multiple parameters and suggests energy replenishment in advance.

[0045] Communication and Interaction: Data is transmitted to the user terminal in real time via Bluetooth. The user terminal interface only displays various parameters and alarms. The system supports voice broadcasting and a built-in vibration motor in the underwear, ensuring that athletes do not need to operate peripherals during exercise. All data is synchronized to the cloud for post-competition analysis.

[0046] In this embodiment, the system achieves multi-dimensional monitoring during a marathon. Compared with traditional single-sensor solutions, the accuracy of vital sign monitoring is significantly improved. For example, relying solely on heart rate can misjudge fatigue as normal load, but by integrating respiratory and hydration data, the system accurately identifies pace decline caused by dehydration and intervenes in a timely manner to prevent athletes from suffering severe dehydration.

[0047] The above embodiments demonstrate the effectiveness and practicality of the present invention in complex sports scenarios. Through multimodal data fusion and intelligent algorithms, the system provides athletes with comprehensive, accurate, and real-time monitoring and feedback.

[0048] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the implementation of the present invention. Those skilled in the art can make other variations or modifications based on the above description. It is neither necessary nor possible to exhaustively describe all embodiments here. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the claims of the present invention.

Claims

1. A method for monitoring motion state, characterized in that, Includes the following steps: S1: Synchronously acquire the user's physiological and motion signals through a multimodal sensor array; S2: Preprocess the collected physiological and motion signals; S3: A deep learning model based on cross-attention mechanism to remove motion artifacts from various physiological signals; S4: Extract multi-dimensional physiological and motion features from the denoised physiological and motion signals; S5: Based on a pre-trained large-scale artificial intelligence model, it fuses and analyzes the extracted physiological and motion features to assess the user's motion status in real time. S6: Generate and output feedback information based on the evaluation results.

2. The motion state monitoring method according to claim 1, characterized in that, Physiological signals include signals acquired by the electrocardiogram sensor (2), photoplethysmography pulse wave sensor (3), lung sound sensor (4), electrical impedance sensor (5), temperature sensor (7), and respiratory wave sensor (8); motion signals include signals acquired by the inertial measurement unit (6).

3. The motion state monitoring method according to claim 2, characterized in that, In step S2, the preprocessing includes: Bandpass filtering was performed on the physiological signals collected by the electrocardiogram sensor (2), photoplethysmography pulse wave sensor (3), respiratory wave sensor (8), and lung sound sensor (4) to retain the effective frequency components; Mean filtering is performed on the physiological signals collected by the impedance sensor (5) and temperature sensor (7) to suppress random noise; The gravity component and attitude drift of the motion signal acquired by the inertial measurement unit (6) are compensated by Kalman filtering.

4. The motion state monitoring method according to claim 3, characterized in that, Step S3, the steps for removing motion artifacts, include: S301: Divide all signals into equal-time sliding window segments to obtain the corresponding signal segments. ; S302: Input the signal segment into the motion artifact extraction model, and extract the most relevant components to the motion signal within the signal segment, i.e., motion artifacts, through the cross-attention mechanism; S303: Subtract the corresponding motion artifact from the signal segment to obtain the denoised signal.

5. The motion state monitoring method according to claim 4, characterized in that, The calculation methods for motion artifact removal include: ； ； ； ； ； ； In the formula, This indicates the signal segment of the inertial measurement unit; Indicates from The motion state features extracted from them are used as the query matrix; and These respectively represent the signal segments The waveform features extracted are used as the key matrix and value matrix; express and The similarity between them; This indicates that the extraction was performed using the cross-attention mechanism. The features of motion artifacts are used as attention weight matrices; Indicates motion artifacts; Indicates the denoised signal; , , and These are specially trained deep networks.

6. The motion state monitoring method according to claim 5, characterized in that, In step S4: The extracted multidimensional physiological characteristics include: heart rate and heart rate variability; blood oxygen saturation; frequency energy percentage of abnormal respiratory sounds, including wheezing and crackles; tissue water content; body temperature; and respiratory rate. The extracted motion features include: step frequency and motion power.

7. The motion state monitoring method according to claim 1, characterized in that, In step S5, the real-time assessment of the user's exercise status includes: assessing exercise performance through cadence and exercise power, and planning exercise plans in real time; assessing cardiovascular system load and cardiac risk through heart rate and heart rate variability; assessing respiratory system load through blood oxygen, respiratory rate, and the proportion of abnormal breath sounds; and assessing the risk of hypothermia or heatstroke through tissue water content and body temperature.

8. The motion state monitoring method according to claim 1, characterized in that, In step S6, the output feedback information includes: data is transmitted to the mobile phone in real time via Bluetooth, and the comprehensive score, risk warning and historical trend are displayed on the mobile phone interface; voice warning and vibration prompt are set.

9. A smart sports bra system for performing the motion state monitoring method according to any one of claims 1 to 8; characterized in that, include: Underwear body (1); A multimodal sensor array is embedded in the elastic fabric layer of the underwear body (1) for synchronously collecting the user's physiological signals and motion data. The multimodal sensor array includes at least an electrocardiogram sensor (2), a photoplethysmography pulse wave sensor (3), a lung sound sensor (4), an electrical impedance sensor (5), an inertial measurement unit (6), a temperature sensor (7), and a respiratory wave sensor (8). The data processing module, connected to the multimodal sensor array, is used to preprocess the collected physiological signals and motion data, and remove motion artifacts from each physiological signal through a cross-attention mechanism based on a deep learning model. The feature extraction module is used to extract multi-dimensional physiological and motion features from physiological signals and motion data after data processing. The analysis and feedback module uses a pre-trained artificial intelligence model to fuse and analyze extracted physiological and motion features, assess the user's motion status in real time, and generate corresponding feedback information.

10. The intelligent sports bra system according to claim 9, characterized in that, The electrocardiogram sensor (2) is a flexible fabric electrode used to collect heart rate variability and arrhythmia characteristics; the photoplethysmography (3) is set on the underwear body (1) at the position corresponding to the sternum and is used to monitor blood oxygen saturation; the lung sound sensor (4) is a piezoelectric film attached to the underwear body (1) at the position corresponding to the chest and is used to capture respiratory sound characteristics. The impedance sensor (5) measures tissue moisture content using low-frequency current; the inertial measurement unit (6) includes a triaxial accelerometer and a gyroscope for tracking motion posture, step frequency, and impact force; the temperature sensor (7) is located on the underarm of the underwear body (1) for monitoring the difference between core body temperature and body surface temperature; the respiratory wave sensor (8) is a strain sensing fiber woven into the underbust position of the underwear body (1) for extracting respiratory frequency and depth.

Images