A wearable sensing student sports health monitoring system

Abstract

This invention relates to the field of intelligent sensors and health monitoring technology, specifically disclosing a wearable sensing student exercise and health monitoring system. The system includes a multimodal sensing module, an edge computing gateway, a cloud analysis platform, and a user interaction terminal. The multimodal sensing module collects electrocardiogram, motion, and pulse wave signals; the edge computing gateway performs local preprocessing and feature extraction; the cloud analysis platform calculates the degree of cardiovascular stress abnormality based on an individualized dynamic baseline and exercise load prediction model, and performs multi-indicator fusion risk assessment by combining heart rate and recovery rate; the user interaction terminal provides early warnings and guidance. This system achieves accurate correlation assessment between exercise load and cardiovascular stress, improving the accuracy and timeliness of risk warnings.

Description

Technical Field

[0001] This invention belongs to the field of intelligent sensor and health monitoring technology, specifically relating to a wearable sensing student sports and health monitoring system. Background Technology

[0002] With the deep integration of the Internet of Things (IoT) and big data technologies, smart health management has become an important development direction in the modern medical and public health fields. This field aims to achieve dynamic assessment of individual health status and early risk warning through continuous, unobtrusive collection and analysis of physiological data. Its core lies in building a complete technology chain from data perception to intelligent decision-making.

[0003] Sports health monitoring for students is a key application area for smart campuses and youth health promotion. The basic goal of this technology is to use wearable sensing devices to collect students' physiological and kinematic parameters in real time during physical activities, and based on this, assess their exercise status, load level, and potential health risks, in order to provide scientific and personalized exercise guidance and safety warnings.

[0004] Current technologies generally employ smart bracelets or watches that integrate accelerometers and optical heart rate sensors, with monitoring metrics largely limited to basic dimensions such as steps, calorie consumption, and average heart rate. This single-dimensional data is insufficient to comprehensively and accurately reflect the overall physiological load during exercise, particularly failing to effectively assess the dynamic relationship between exercise intensity and cardiovascular stress response. For example, under the same exercise duration, different exercise types or individual differences can lead to drastically different cardiac loads, and current systems lack simultaneous analysis and integration of in-depth physiological indicators such as heart rate variability and post-exercise heart rate recovery rate. This limitation in monitoring dimensions makes it difficult for the system to identify hidden risks of overexertion or early signs of abnormal cardiac function, easily leading to missed detections of potential health threats. Furthermore, existing solutions typically use static thresholds for early warning, failing to combine individual student baseline data with real-time exercise context for dynamic risk assessment, resulting in insufficient accuracy and timeliness of warnings.

[0005] Therefore, how to construct a student exercise health monitoring system that can integrate multi-dimensional physiological and exercise data and achieve accurate assessment of the dynamic correlation between exercise load and cardiovascular risk has become an urgent technical challenge. Summary of the Invention

[0006] This invention provides a wearable sensing student sports and health monitoring system to solve the problems of existing technologies, such as single monitoring dimensions, inability to accurately correlate exercise load with cardiovascular stress response, and insufficient accuracy and timeliness of early warnings due to static risk assessment.

[0007] The technical solution of the present invention is a wearable sensing student sports and health monitoring system, which includes a multimodal sensing module, an edge computing gateway, a cloud analysis platform, and a user interaction terminal.

[0008] The multimodal sensing module is used to collect multi-dimensional physiological and kinematic raw data during student movement without physical contact. The module consists of multiple distributed miniature sensor nodes worn on specific parts of the student's body. Specifically, it includes an ECG sensing node integrated on a chest strap or patch base for collecting high-fidelity ECG signals; a triaxial accelerometer and gyroscope node integrated in a wrist device for collecting limb acceleration and angular velocity data; and an optical volumetric imaging sensing node integrated in the same wrist device for collecting photoplethysmography (PPI) signals. All sensor nodes establish wireless data connections with an edge computing gateway via Bluetooth Low Energy.

[0009] The edge computing gateway is used to perform local preprocessing, feature extraction, and preliminary fusion of the raw data collected by the multimodal sensing module. Deployed within the campus local network environment, the edge computing gateway runs an edge computing engine. The edge computing engine first performs adaptive filtering on the raw electrocardiogram signal to eliminate electromyographic interference, and then uses real-time... The wave detection algorithm extracts the beat-by-beat heart rate sequence, and then calculates the time-domain and frequency-domain heart rate variability indices based on the beat-by-beat heart rate sequence.

[0010] Simultaneously, the edge computing engine performs integral and vector synthesis operations on the acceleration and angular velocity data to calculate the exercise intensity index and exercise type recognition features. Further, the edge computing engine performs waveform reconstruction and feature point detection on the optical volumetric pulse wave signal, extracting pulse wave conduction time as an auxiliary indicator of cardiovascular system stress. After completing the above feature extraction, the edge computing engine performs the first-level data fusion, aligning the heart rate variability index, exercise intensity index, exercise type features, and pulse wave conduction time by timestamp, encapsulating them into structured data frames, and uploading them to the cloud analysis platform in real time via an encrypted network link.

[0011] The cloud-based analytics platform receives structured data frames uploaded from the edge computing gateway and performs individualized modeling, dynamic risk assessment, and decision generation. The core of this platform is a dynamic assessment model of exercise cardiovascular load. The construction and operation of this model are as follows: First, the platform establishes an independent personal physiological profile for each student, storing their historical resting heart rate, historical heart rate variability baseline, age, gender, and body mass index. Upon receiving a real-time data frame, the model activates the individualized dynamic baseline calculation unit. This unit dynamically generates a personalized heart rate variability reference baseline suitable for the current exercise situation based on the student's heart rate variability data from the five minutes prior to exercise, combined with their historical resting baseline, using a sliding window weighted average algorithm.

[0012] Subsequently, the load risk correlation analysis unit in the model begins operation. This unit takes the real-time exercise intensity index as input and, through a pre-trained neural network mapping function, predicts the theoretical deviation of the heart rate variability index (HRV) from its personalized dynamic baseline at the current exercise intensity. This HRV mapping function is obtained through supervised learning training on a large historical dataset labeled with exercise type and intensity, and corresponding HRV changes. The load risk correlation analysis unit then calculates the residual between the actual measured value of the HRV and the aforementioned theoretical prediction. This residual is defined as the cardiovascular stress abnormality.

[0013] Meanwhile, the model's built-in risk decision-making logic unit continuously monitors three key indicators: whether the real-time heart rate exceeds 85% of the maximum predicted heart rate based on age, whether the cardiovascular stress abnormality exceeds three consecutive data sampling periods with an amplitude greater than a preset threshold, and whether the heart rate recovery rate within three minutes after exercise is lower than 70% of the individual's historical average recovery rate. The risk decision-making logic unit performs logical and weighted judgments on the above indicators. When any indicator is triggered, or when multiple indicators exhibit a combined abnormal pattern, the risk decision-making logic unit generates risk warning instructions of different levels and corresponding exercise intervention suggestions.

[0014] The user interaction terminal is used to receive and visualize risk assessment results, early warning information, and exercise guidance suggestions issued by the cloud-based analysis platform. This user interaction terminal is a dedicated application installed on the teacher's or school doctor's management device. The application interface displays a real-time list of students, their current exercise status, heart rate, and workload level. When an early warning command is received, the terminal pushes out warning details with highlighted colors and sound prompts, and displays suggested interventions, such as reducing exercise intensity, pausing exercise, or undergoing a medical examination.

[0015] As one embodiment of the present invention, the sliding window weighted average algorithm of the individualized dynamic baseline calculation unit in the exercise cardiovascular load dynamic assessment model is specifically implemented as follows: the high-frequency power spectral density value of heart rate variability in the 5 minutes before the start of exercise is selected as the time series, the window length is set to 2 minutes, and the step size is 30 seconds; time decay weights are assigned to the data in the window, and the data closer to the start of exercise has a higher weight; the weighted average value of each window is calculated, and the weighted average value of the last window is set as the individualized dynamic baseline of this exercise.

[0016] In one embodiment of the present invention, the pre-trained neural network mapping function in the load risk correlation analysis unit has a fully connected feedforward network structure comprising one input layer, two hidden layers, and one output layer. The input layer neurons receive the normalized real-time motion intensity index and motion type encoding. The first hidden layer contains 64 neurons and employs a rectified linear unit activation function. The second hidden layer contains 32 neurons, also employing a rectified linear unit activation function. The output layer consists of one neuron, outputting the logarithmic transformation value of the predicted heart rate variability high-frequency power spectral density. The network is trained using mean squared error as the loss function and optimized using an adaptive moment estimation algorithm.

[0017] In one embodiment of the present invention, the combined abnormal pattern recognition logic in the risk decision-making logic unit specifically defines two combined abnormal patterns. The first pattern is when the real-time heart rate exceeds 85% of the maximum predicted heart rate, and the cardiovascular stress abnormality is positive for two consecutive cycles. The second pattern is when the heart rate recovery rate after exercise is lower than 70% of the historical average, and the cardiovascular stress abnormality immediately after exercise stops exceeds twice the preset threshold. When the first pattern is identified, a high-risk red alert is generated, and it is recommended to stop exercising immediately. When the second pattern is identified, a medium-risk yellow alert is generated, and close observation and subsequent heart rate recovery tracking are recommended.

[0018] In one embodiment of the present invention, the edge computing gateway also integrates a local caching and disconnection resumption mechanism. When a network connection interruption with the cloud analytics platform is detected, the edge computing engine temporarily stores the processed structured data frames in local non-volatile memory. After the network connection is restored, the gateway automatically retransmits the cached data frames to the cloud in chronological order, ensuring data continuity.

[0019] In one embodiment of the present invention, the system operates within a hierarchical data processing architecture. A multimodal sensing module is responsible for millisecond-level raw data acquisition. An edge computing gateway is responsible for second-level feature extraction and data compression. A cloud-based analytics platform is responsible for minute-level model computation and risk assessment. A user interaction terminal is responsible for second-level result presentation and alarms. This hierarchical data processing architecture effectively balances system real-time performance, computational load, and energy consumption constraints.

[0020] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention achieves a deep correlation and precise quantitative assessment of exercise load and cardiovascular stress response by constructing a multimodal sensing module and a dynamic assessment model of exercise cardiovascular load. The system not only collects baseline heart rate but also simultaneously analyzes in-depth physiological indicators such as heart rate variability and pulse wave transit time, and performs dynamic modeling in conjunction with exercise intensity. By calculating the core indicator of cardiovascular stress abnormality, the system can effectively identify hidden cardiac overload or functional abnormalities that may be masked by conventional heart rate monitoring, significantly improving the early identification and assessment accuracy of exercise-related health risks, especially cardiogenic risks.

[0021] 2. This invention abandons the static, universal threshold-based early warning method and introduces an individualized dynamic baseline calculation mechanism based on personal historical data and real-time pre-exercise state. This individualized dynamic baseline calculation mechanism allows the risk assessment reference standard to adapt to the student's daily physiological fluctuations and current physical state, avoiding false alarms and missed alarms caused by individual differences or changes in state. Simultaneously, the risk decision-making logic unit employs a multi-indicator fusion and combined abnormal pattern recognition strategy, comprehensively considering multi-dimensional evidence such as heart rate, load correlation anomaly degree, and recovery ability, making the final risk judgment more comprehensive and robust, greatly enhancing the accuracy and reliability of the early warning signal.

[0022] 3. This invention employs a layered computing architecture that coordinates edge and cloud computing, offloading computationally complex feature extraction and data preprocessing tasks to the edge computing gateway, while only uploading lightweight structured feature data to the cloud. This design significantly reduces the bandwidth requirements and power consumption of wireless data transmission, ensuring the long-term battery life of wearable sensor nodes. Simultaneously, the cloud focuses on complex individualized modeling and risk assessment, fully utilizing its powerful storage and computing resources. This layered computing architecture optimizes resource allocation and enhances the system's practicality and scalability while ensuring overall system performance and advanced functionality. Attached Figure Description

[0023] Figure 1 This is a schematic diagram of the overall technical solution architecture of the present invention; Figure 2 This is a schematic diagram of the core principle framework of the dynamic assessment model of exercise cardiovascular load in this invention; Figure 3 This is a logical flowchart of the edge computing gateway performing data preprocessing and feature extraction in this invention; Figure 4 This is a schematic diagram of the multi-level interaction relationship and data flow between the multimodal sensing module, the edge computing gateway, and the cloud analysis platform in this invention; Figure 5 This is a flowchart illustrating the logical process of multi-indicator fusion judgment within the risk decision-making logic unit of this invention. (Detailed implementation details follow.) Example 1: Please refer to the appendix Figure 1 To be continued Figure 5 This invention provides a wearable sensing student sports and health monitoring system. The system consists of four core components: a multimodal sensing module, an edge computing gateway, a cloud analysis platform, and a user interaction terminal. These components communicate with each other via standardized communication protocols and data interfaces to achieve highly reliable, low-latency information exchange. The entire system operates according to a layered processing principle: the multimodal sensing module is responsible for acquiring millisecond-level raw physiological and kinematic signals; the edge computing gateway performs second-level local feature extraction, data fusion, and compression; the cloud analysis platform undertakes minute-level individualized modeling and dynamic risk assessment; and the user interaction terminal provides second-level response times for visualizing early warning information and pushing intervention commands. This layered architecture effectively balances the contradictions between real-time performance, computational load, energy consumption constraints, and functional integrity, ensuring stable and efficient operation of the system in the complex network environment of a campus.

[0024] The multimodal sensing module, as the data sensing unit at the system's front end, employs a distributed micro-sensor node layout strategy, deploying it at key anatomical locations on the student's body to achieve seamless, continuous, and high-fidelity multi-dimensional data acquisition. Specifically, the multimodal sensing module comprises three functionally independent but time-synchronized sensing sub-units: an ECG sensing node, an inertial motion sensing node, and an optical volumetric plethysmography sensing node. The ECG sensing node is integrated onto a flexible chest strap or disposable medical patch base, closely adhering to the student's chest skin. It acquires standard lead I ECG signals via two dry electrodes, with a sampling frequency set at 500 Hz and a resolution of at least 12 bits to ensure... Precise capture of wave morphology details and microvolt-level ST segment changes.

[0025] Inertial motion sensing nodes and optical volumetric plethysmography (OPL) sensing nodes are integrated into a single wrist-worn wearable device. The inertial motion sensing node includes a three-axis accelerometer and a three-axis gyroscope, synchronously recording the linear acceleration and angular velocity of the wrist in three-dimensional space at sampling rates of 100 Hz and 200 Hz, respectively. The OPL sensing node uses dual-wavelength (typically 660 nm red light and 940 nm infrared light) LEDs in conjunction with a photodetector to acquire transmitted or reflected photoplethysmography (PPP) signals at a sampling rate of 250 Hz. All sensing nodes are equipped with a real-time clock chip, and cross-node time synchronization is achieved through a hardware triggering mechanism, with time alignment errors controlled within ±2 milliseconds. Raw data is periodically transmitted to the edge computing gateway via Bluetooth Low Energy 5.0 in broadcast mode at 1-second intervals. Each transmission contains a complete data packet, including a timestamp, sensor type identifier, raw sampling sequence, and checksum. To ensure data security, all wireless transmissions are encrypted end-to-end using the AES-128 encryption algorithm.

[0026] The edge computing gateway, serving as the core hub for local intelligent processing within the system, is deployed in fixed locations within the campus LAN, such as sports equipment rooms, medical rooms, or computer labs in teaching buildings. Its hardware platform is based on an embedded ARM Cortex-A series processor, equipped with at least 512 megabytes of RAM and 8 gigabytes of non-volatile storage, and includes a built-in network interface supporting Bluetooth Low Energy and Ethernet / Wi-Fi dual-mode communication. Internally, the gateway runs a lightweight Linux operating system, on which an edge computing engine is deployed. This edge computing engine is based on... Figure 3 The logical flow framework shown performs data preprocessing, feature extraction, and preliminary fusion tasks. Upon receiving the raw data packets uploaded by the multimodal sensing module, the edge computing engine first aligns and reassembles the data from different sensing nodes according to the timestamps, forming multi-channel synchronous data blocks in 1-second increments.

[0027] For the ECG signal, the edge computing engine initiates an adaptive filtering subroutine. This subroutine employs a recursive filtering algorithm based on the minimum mean square error criterion to dynamically estimate and suppress electromyographic interference noise caused by limb movement. The filtered signal is then input to the real-time... Wave detection module, this real-time The wave detection module implements frame-by-frame detection based on an improved Pan-Tompkins algorithm. Wave localization, the core steps of which include: squaring the filtered signal to enhance the QRS group energy, smoothing the waveform through sliding window integration, and finally setting a dynamic threshold for peak detection. Interwave sequence is thus generated and used to calculate heart rate variability indices in the time and frequency domains. The time-domain indices include adjacent normal values. The root mean square value of the difference between intervals ( The calculation formula is as follows: in Valid detected within the current 1-second window Number of intervals. Frequency domain metrics, on the other hand, are determined by... The interval sequence was interpolated and resampled to 4 Hz, and then the power spectral density was calculated using Fast Fourier Transform, with a focus on extracting the power values ​​in the high-frequency band (0.15–0.4 Hz). This frequency domain index reflects the regulatory activity of the parasympathetic nervous system on the heart.

[0028] For inertial motion data, the edge computing engine first performs zero-bias correction and unit conversion on the three-axis acceleration and three-axis angular velocity, respectively, and then performs vector synthesis calculations. Motion intensity index ( The acceleration vector magnitude is calculated as follows: the absolute value of the acceleration vector magnitude minus the static gravity component is taken, and then a weighted sum is performed with the angular velocity vector magnitude. The weighting coefficients are determined to be 0.7 and 0.3 based on a large number of adolescent sports samples. Simultaneously, based on the time-frequency characteristics of acceleration and angular velocity (such as zero-crossing rate, energy entropy, and dominant frequency components), the engine calls a pre-trained lightweight support vector machine classifier to identify the current sports type and outputs a category label encoded as an integer (e.g., 0 for sitting, 1 for walking, 2 for jogging, 3 for high-intensity interval running, etc.).

[0029] After the optical volumetric imaging signal is bandpass filtered to remove the DC component and high-frequency noise, it enters the waveform reconstruction stage. The engine uses cubic spline interpolation to smooth the signal and employs a combined first- and second-order derivative detection method to accurately locate the foot point and systolic peak of the pulse wave. Pulse wave conduction time (PWTT) is defined as the intracardiac electrical activity within the same cardiac cycle. The time difference between the peak value and the onset of the pulse wave, measured in milliseconds. This frequency domain index has been shown to be negatively correlated with arterial blood pressure and can serve as a sensitive auxiliary parameter for acute stress response in the cardiovascular system.

[0030] After completing the feature extraction described above, the edge computing engine performs the first-level data fusion. , Exercise intensity index, exercise type code, PWTT and real-time heart rate (from (Calculated using the reciprocal of the interval) and encapsulated into structured data frames with a unified timestamp. Each frame contains 12 fields: timestamp (Unix millisecond time), student's unique identifier, and real-time heart rate (beats / minute). (millisecond), The data frame contains the following parameters: milliseconds², exercise intensity index (dimensionless), exercise type code (integer), PWTT (milliseconds), ECG signal quality index (0–100), optical signal quality index (0–100), acceleration signal quality index (0–100), and data integrity flag. After Base64 encoding and HMAC-SHA256 signing, this structured data frame is uploaded to the cloud analysis platform in real time via HTTPS. If a network connection interruption is detected, the gateway immediately activates its local caching mechanism, writing the data frame to be transmitted into a circular buffer in non-volatile memory. The maximum cache capacity supports continuous offline data storage for 72 hours. After the network is restored, the gateway automatically retransmits all cached frames in chronological order and sends a data continuity verification request to the cloud to ensure no historical records are missing.

[0031] The cloud-based analytics platform, serving as the system's intelligent decision-making hub, is deployed on a cloud server cluster with high availability and elastic scalability. Its core is a dynamic assessment model for exercise cardiovascular load. The operational logic of this assessment model is shown in the attached diagram. Figure 2 As shown, the platform comprises three main functional modules: an individualized dynamic baseline calculation unit, a load-risk correlation analysis unit, and a risk decision-making logic unit. The platform creates a unique personal physiological profile for each registered student, including but not limited to: date of birth, gender, height, weight (for calculating body mass index), a summary of the ECG from the entrance physical examination, and the average heart rate at rest for the past 30 days. and Baseline values. Archive data is automatically updated daily at midnight, and a version control mechanism is set up to track long-term evolution trends in physiological states.

[0032] When the cloud receives a structured data frame uploaded by the edge computing gateway, it first parses the student's identity and loads their personal physiological profile. If the current data frame corresponds to a newly started exercise session (i.e., the timestamp interval of the previous frame exceeds 5 minutes), the individualized dynamic baseline calculation unit is triggered. This individualized dynamic baseline calculation unit reviews all resting or low-intensity activity data frames uploaded by the student within 5 minutes before the start of this exercise, filters out data points with a heart rate below 100 beats / minute and an exercise intensity index less than 0.3, and extracts data from them. The values ​​constitute a time series. Subsequently, a sliding window weighted average algorithm is applied to generate a personalized dynamic baseline. Specifically, the window length is set to 120 seconds, the step size to 30 seconds, and a total of 8 overlapping windows are generated; for each window... The value is assigned an exponentially decaying weight, and the weighting function is defined as follows: ,in The moment the exercise begins. Attenuation coefficient at the time of data point acquisition Set the value to 0.02; calculate the weighted average of each window, and finally take the weighted average of the last window (i.e., the window closest to the start of the motion) as the total weighted average of the entire motion. Dynamic reference baseline .

[0033] The load risk correlation analysis unit then starts working. This individualized dynamic baseline calculation unit receives the motion intensity index from the current data frame. With motion type coding After normalization, it is input into the pre-trained neural network mapping function. Predicting the current exercise load Theoretical expected value This neural network is a fully connected feedforward structure, containing one input layer (two neurons: normalized). The system consists of one-hot encoded motion type, two hidden layers (64 and 32 neurons respectively, both using rectified linear unit activation functions), and one output layer (1 neuron, outputting...). The online training phase used a historical database containing 100,000 students and a total of 5 million hours of labeled data, with labels based on actual observations. The loss function is the mean squared error, the optimizer is the adaptive moment estimation algorithm, and the initial learning rate is set to 0.001. During the inference phase, the network output is obtained after an exponential transformation. Cardiovascular stress abnormality Defined as actual measured value Compared with theoretical predictions The residual, i.e. A positive value indicates that the degree of parasympathetic inhibition is greater than expected, suggesting possible hidden cardiac overload or autonomic nervous system dysfunction.

[0034] The risk decision-making logic unit continuously monitors three key indicators, and its judgment logic is as follows: Figure 5 As shown. The first item is real-time heart rate exceedance: calculate the student's maximum predicted heart rate. If the current heart rate If the heart rate exceeds the limit, it is marked as an event exceeding the heart rate limit. The second item is a persistent exceedance of cardiovascular stress abnormality: if... (A preset threshold, typically 1.5 times the individual's historical standard deviation) and if this state is maintained for 3 or more consecutive data sampling periods (i.e., 3 seconds), an anomaly alarm is triggered. The third item is decreased heart rate recovery ability: when the system detects an exercise cessation event (defined as an exercise intensity index below 0.1 for 10 consecutive seconds), heart rate recovery monitoring is activated. The heart rate at the end of the 3rd minute after exercise cessation is calculated. And compared with the student's average heart rate recovery rate 3 minutes after exercise in history. Comparison. If If so, it is determined that the recovery ability has significantly decreased.

[0035] The risk decision-making logic unit not only independently evaluates the above three indicators, but also performs combined abnormal pattern identification. The system predefines two high-risk combination patterns: Pattern 1 is "heart rate exceeding limits + persistently positive abnormality value," i.e. and Two consecutive cycles; Mode 2 is "decreased recovery ability + a sharp increase in abnormality immediately after exercise termination", that is, the heart rate recovery rate is lower than 70% of the historical average and the time of exercise cessation. When Mode 1 is detected, the system generates a Level 1 red alert, recommending immediate cessation of exercise and notification of the school doctor; when Mode 2 is detected, a Level 2 yellow alert is generated, recommending suspension of further strenuous activity and scheduling a follow-up electrocardiogram (ECG) within 24 hours. Furthermore, if a single indicator is severely abnormal (e.g., heart rate exceeding a certain threshold), further warnings will be issued. ,or (This can also directly trigger a red alert.)

[0036] The user interaction terminal is a dedicated application installed on teachers' tablets or school doctor's workstations. Its interface design follows medical human factors engineering principles to ensure clear information delivery and simple operation. The main interface displays all students currently participating in monitoring in a list format. Each row includes the student's name, avatar, real-time heart rate, exercise intensity level (distinguished by color: green for low, yellow for medium, and red for high), and the latest risk status icon.

[0037] When the cloud analytics platform issues an alert, a full-screen alert window immediately pops up at the top of the terminal screen. The background color is set to red or yellow depending on the alert level, accompanied by a specific frequency audio prompt (a rapid double tone for a red alert, and a single long tone for a yellow alert). The alert details page clearly lists the triggering indicators, values, reference thresholds, and system-suggested interventions, such as "It is recommended to reduce running speed to walking level" or "Please go to the medical room for an electrocardiogram." All alert events are automatically recorded in the cloud log and a structured report is generated for subsequent health record archiving and epidemiological analysis.

[0038] The entire system is connected via an attachment Figure 4 The multi-level interaction relationships shown achieve a data closed loop. The multimodal sensing module generates massive amounts of data at a raw sampling rate of 1000 Hz, which is then compressed into structured feature data of 1 frame per second by the edge computing gateway, reducing the data volume by more than 99% and greatly reducing the burden on wireless transmission. The cloud platform performs deep modeling based on minute-level aggregated data, and its output risk assessment results guide on-site teaching and health management decisions through user interaction terminals. This closed-loop mechanism ensures that the system not only has high-precision monitoring capabilities but also forms a complete health management chain of "perception—analysis—decision—intervention—feedback".

[0039] Example 2: Building upon the previous examples, this example further enhances the local caching and resuming-from-disconnect mechanisms of the edge computing gateway to address common network fluctuations and signal blind spots in campus environments. The edge computing gateway's non-volatile memory is divided into two independent areas: a main cache and a verification log area. The main cache uses a circular queue structure with a maximum capacity of 8 gigabytes, storing unsuccessfully uploaded structured data frames according to a first-in, first-out (FIFO) principle. Each data frame is appended with a unique sequence number and hash checksum upon writing. The verification log area records the time, target address, return status code, and number of retries for each upload attempt. Log entries are appended to ensure no data loss during power outages.

[0040] When the gateway detects a TCP connection timeout (default threshold of 10 seconds) or an HTTP response code of 5xx / 4xx with the cloud analytics platform, it immediately switches to offline mode. In this mode, the edge computing engine continues to perform data acquisition, feature extraction, and fusion tasks normally, but temporarily stores the output data frames in the main buffer instead of uploading them directly. Simultaneously, the gateway starts a background heartbeat detection thread, attempting to establish a lightweight connection with the cloud every 30 seconds (sending only PING packets). Once the connection is restored, the gateway enters the retransmission phase. The retransmission process uses a chunked transmission strategy: each time, the earliest 100 frames of data in the main buffer are read, packaged into a batch upload request, and sent via the HTTPS POST method. After receiving the batch request, the cloud first verifies the hash checksum of each frame, and then checks the data continuity by sequence number. If a missing frame is found (e.g., a skipped sequence number), the cloud proactively sends a retransmission request to the gateway, specifying the sequence number range of the missing frame. Upon receiving this request, the gateway prioritizes retrieving and retransmitting the specified frame from the main buffer. After all retransmissions are completed, the gateway clears the confirmed received frames from the main buffer and updates the verification log.

[0041] Furthermore, this embodiment introduces a data prioritization mechanism. In the main buffer, data frames with warning flags (i.e., those determined to be abnormal by the risk decision-making logic unit) are assigned high priority and stored at the front of the queue. In the initial retransmission after network recovery, the system prioritizes uploading all high-priority frames to ensure that critical health event information can be delivered to the cloud and user terminals as quickly as possible. Ordinary data frames are then retransmitted in the normal order. This mechanism significantly improves the system's clinical response capability in weak network environments.

[0042] Example 3: This example focuses on the online learning and adaptive evolution capabilities of a dynamic assessment model of exercise cardiovascular load in a cloud-based analytics platform. Traditional models, once deployed, remain fixed and cannot adapt to seasonal changes in the physiological characteristics of the student population or long-term drift caused by individual development. Therefore, this example adds an incremental model update module to the cloud platform. The module automatically executes the following process every morning: First, it selects samples marked as "high-quality" (defined as data frames with a signal quality index greater than 80 and no warning events) from all successfully uploaded structured data frames in the past 24 hours; second, it mixes these new samples with the historical training set at a ratio of 1:10 to form an incremental training set; then, it calls the distributed training framework to map the neural network function. Fine-tuning is performed, with the learning rate decayed to 0.0001 to avoid catastrophic forgetting. Finally, the generalization performance of the fine-tuned model is evaluated using a reserved validation set. If the mean squared error decreases by more than 0.5%, the new model version is deployed to the production environment, and the model reference version number in the personal physiological profiles of all students is updated.

[0043] Meanwhile, the individualized dynamic baseline calculation unit incorporates a long-term trend correction factor. The system calculates the baseline for each student over the past 7 days. Moving average of dynamic baseline If the currently generated and If the relative deviation exceeds 15%, a manual review process is triggered, notifying the school doctor to check for potential interfering factors such as recent colds, lack of sleep, or psychological stress in the student. This mechanism effectively prevents the model from making misjudgments due to short-term physiological fluctuations, enhancing the system's robustness and reliability in complex real-world scenarios.

Claims

1. A wearable sensing student physical health monitoring system, characterized in that, include: The multimodal sensing module is used to collect multi-dimensional physiological and kinematic raw data of students during exercise without physical contact. Edge computing gateways are used to perform local preprocessing, feature extraction, and preliminary fusion of raw data collected by multimodal sensing modules; The cloud-based analytics platform receives structured data frames uploaded from the edge computing gateway and performs individualized modeling, dynamic risk assessment, and decision generation. The user interaction terminal is used to receive and visualize risk assessment results, early warning information and exercise guidance suggestions issued by the cloud analysis platform; The user interaction terminal is a dedicated application installed on the management equipment of teachers or school doctors; The application interface displays the student list, current exercise status, heart rate, and load level in real time, and pushes warning details and suggested intervention measures with highlighted colors and sound prompts when it receives a warning command.

2. The wearable sensing student sports health monitoring system as claimed in claim 1, wherein, The multimodal sensing module includes multiple miniature sensor nodes distributed and worn on specific parts of the student's body; The multiple micro-sensor nodes include ECG sensing nodes integrated on a chest strap or patch base, triaxial accelerometer and gyroscope nodes integrated in a wrist device, and optical volumetric sensing nodes integrated in the same wrist device. All sensor nodes establish wireless data connections with the edge computing gateway via Bluetooth Low Energy protocol. The edge computing gateway runs an edge computing engine internally; The edge computing engine is used for adaptive filtering of the original electrocardio signal to eliminate electromyographic interference, and real-time Wave detection algorithm extracts beat-to-beat heart rate sequence, and then calculates time domain and frequency domain heart rate variability indexes based on the beat-to-beat heart rate sequence.

3. The wearable sensor based student physical health monitoring system as claimed in claim 1, wherein, The edge computing engine is used to perform integral and vector synthesis operations on acceleration and angular velocity data, and to calculate motion intensity index and motion type identification features; Furthermore, the edge computing engine is used to reconstruct the waveform and detect feature points of the optical volumetric pulse wave signal, and extract the pulse wave conduction time; After completing the above feature extraction, the edge computing engine performs the first-level data fusion, aligning the heart rate variability index, exercise intensity index, exercise type features, and pulse wave conduction time by timestamp, encapsulating them into structured data frames, and uploading them to the cloud analysis platform in real time through an encrypted network link.

4. The wearable sensor based student physical health monitoring system as claimed in claim 1, wherein, The core of the cloud-based analysis platform is a dynamic assessment model for exercise cardiovascular load. The exercise cardiovascular load dynamic assessment model includes an individualized dynamic baseline calculation unit, a load risk correlation analysis unit, and a risk decision logic unit; The individualized dynamic baseline calculation unit is used to dynamically generate a personalized heart rate variability reference baseline suitable for the current exercise situation based on the student's heart rate variability data in the 5 minutes before the current exercise, combined with their historical resting baseline, and through a sliding window weighted average algorithm. The load risk correlation analysis unit is used to take the real-time exercise intensity index as input, and through a pre-trained neural network mapping function, predict the theoretical deviation of the heart rate variability index relative to its personalized dynamic baseline under the current exercise intensity, and calculate the residual between the actual measured value and the theoretical predicted value of the heart rate variability index. This residual is defined as the degree of cardiovascular stress abnormality. The risk decision-making logic unit is used to continuously monitor whether the real-time heart rate exceeds 85% of the maximum predicted heart rate based on age, whether the cardiovascular stress abnormality exceeds 3 consecutive data sampling cycles and the amplitude is greater than the preset threshold, and whether the recovery rate of the heart rate within 3 minutes after stopping exercise is lower than 70% of the average historical recovery rate of the individual. The unit performs logical and weighted judgment on the above indicators. When any one indicator is triggered or multiple indicators show a combined abnormal pattern, risk warning instructions of different levels and corresponding exercise intervention suggestions are generated.

5. The wearable sensor based student physical health monitoring system as claimed in claim 1, wherein, The specific implementation process of the sliding window weighted average algorithm in the individualized dynamic baseline calculation unit is as follows: the high-frequency power spectral density value of heart rate variability 5 minutes before the start of exercise is selected as the time series, the window length is set to 2 minutes, and the step size is 30 seconds; the data in the window are assigned time decay weights, and the data closer to the start of exercise has a higher weight; Calculate the weighted average for each window and set the weighted average of the last window as the personalized dynamic baseline for this movement.

6. The wearable sensor based student physical health monitoring system as claimed in claim 1, wherein, The pre-trained neural network mapping function in the load risk correlation analysis unit has a network structure of a fully connected feedforward network containing one input layer, two hidden layers, and one output layer; the input layer neurons receive the normalized real-time motion intensity index and motion type code. The first hidden layer contains 64 neurons and uses the rectified linear unit activation function; The second hidden layer contains 32 neurons and also uses the rectified linear unit activation function; The output layer consists of one neuron, which outputs the logarithmic transform of the predicted high-frequency power spectral density value of heart rate variability. The neural network is trained using mean squared error as the loss function and optimized using an adaptive moment estimation algorithm.

7. The wearable sensor based student physical health monitoring system as claimed in claim 1, wherein, The combined anomaly pattern recognition logic in the risk decision-making logic unit specifically involves: the system defining two combined anomaly patterns; The first mode is when the real-time heart rate exceeds the threshold of 85% of the maximum predicted heart rate, and the cardiovascular stress abnormality is positive for two consecutive cycles. The second mode is when the heart rate recovery rate after exercise is less than 70% of the historical average, and the cardiovascular stress abnormality immediately after exercise stops exceeds twice the preset threshold. When the first pattern is detected, a high-risk red alert is generated and it is recommended to stop exercising immediately. When the second pattern is identified, a medium-risk yellow alert is generated, and close observation and follow-up heart rate recovery are recommended.

8. A wearable sensing student sports and health monitoring system according to claim 1, characterized in that, The edge computing gateway also integrates local caching and a mechanism for resuming transmission after disconnection; When a network connection interruption with the cloud analytics platform is detected, the edge computing engine temporarily stores the processed structured data frames in local non-volatile memory. After the network connection is restored, the edge computing gateway automatically retransmits the cached data frames to the cloud analysis platform in chronological order. The local caching and disconnection resumption mechanism also includes a data priority mechanism; in the local cache, data frames with warning flags are given high priority and are uploaded first after the network is restored.

9. A wearable sensing student sports and health monitoring system according to claim 1, characterized in that, The process by which the edge computing engine calculates the time-domain heart rate variability index includes calculating adjacent normal values. The root mean square value of the difference between intervals ( The calculation formula is as follows: . The dynamic assessment model of exercise cardiovascular load also includes a model incremental update module; The incremental model update module is used to periodically filter samples from newly uploaded high-quality data, fine-tune the pre-trained neural network mapping function, and deploy a new model version when the model performance is improved by more than 0.5%.

10. A wearable sensing student sports and health monitoring system according to claim 1, characterized in that, The individualized dynamic baseline calculation unit also incorporates a long-term trend correction factor. The system calculates the moving average of each student's dynamic baseline over the past 7 days. If the relative deviation between the currently generated dynamic baseline and this moving average exceeds 15%, a manual review process is triggered. The system operates on a hierarchical data processing architecture; The multimodal sensing module is responsible for acquiring raw data at the millisecond level; The edge computing gateway is responsible for second-level feature extraction and data compression; The cloud-based analytics platform is responsible for minute-level model calculations and risk assessments. The user interaction terminal is responsible for providing results and issuing alerts within seconds.

Images