Exercise guidance method and equipment based on health detection

Abstract

The invention discloses an exercise guidance method and equipment based on health detection, and relates to the field of health detection.The method comprises the steps that physiological data, behavior data and environment data of a user are collected, and the physiological data, the behavior data and the environment data are preprocessed to generate a dynamic health file of the user; inputting the dynamic health archive into the lightweight deep neural network model, optimizing the lightweight deep neural network model based on the dynamic health archive by adopting a self-supervised learning and anomaly detection algorithm, and generating a real-time health state of the user by utilizing the optimization model; obtaining health trend prediction data issued by the cloud server, wherein the health trend prediction data is generated based on historical health data of a user uploaded by a terminal device; and in combination with the real-time health state and health trend prediction data of the user, a self-adaptive motion guidance strategy is generated through a reinforcement learning algorithm. Through multi-dimensional data fusion and algorithm optimization, the personalized ability of health assessment and exercise guidance is improved.

Description

Technical Field

[0001] The present application relates to the field of health detection, and in particular to a method and device for guiding exercise based on health detection. Background Art

[0002] Currently, smart wearable devices have been widely used in the fields of health monitoring and exercise guidance. They mainly collect physiological data such as user heart rate, blood oxygen, and exercise status through photoplethysmography and acceleration sensors, and generate health assessment results and exercise recommendations based on preset rules or general algorithms.

[0003] However, most existing solutions are designed as static algorithm models that rely on limited parameters such as age and gender. They are unable to dynamically adjust to individual users' physiological differences, behavioral habits, and environmental changes, resulting in insufficient accuracy in health monitoring. Specifically, fixed-threshold heart rate anomaly detection is difficult to adapt to the physical fitness baselines of different users, while unified exercise recommendation strategies ignore users' long-term health trends and real-time status changes. In addition, traditional methods lack the ability to collaboratively analyze multi-dimensional data, further limiting the level of personalization of health assessments and exercise guidance. Summary of the Invention

[0004] The main purpose of this application is to provide a method and device for exercise guidance based on health detection, aiming to improve the personalized ability of health assessment and exercise guidance for users.

[0005] To achieve the above objectives, the present application provides a health detection-based exercise guidance method applied to a terminal device, the health detection-based exercise guidance method comprising:

[0006] Collecting the user's physiological data, behavioral data and environmental data, and pre-processing the physiological data, behavioral data and environmental data to generate a dynamic health profile of the user;

[0007] Inputting the dynamic health record into a lightweight deep neural network model, optimizing the lightweight deep neural network model based on the dynamic health record using a self-supervised learning and anomaly detection algorithm, and generating the user's real-time health status using the optimized model;

[0008] Obtaining health trend prediction data issued by a cloud server, where the health trend prediction data is generated based on the user's historical health data uploaded by the terminal device;

[0009] Combining the user's real-time health status and health trend prediction data, an adaptive exercise guidance strategy is generated through a reinforcement learning algorithm.

[0010] In one embodiment, the step of optimizing the lightweight deep neural network model based on the dynamic health record using self-supervised learning and anomaly detection algorithms includes:

[0011] Based on the dynamic health profile, construct a self-supervised training task;

[0012] Updating health detection parameters of the lightweight deep neural network model using the loss function of the self-supervised training task;

[0013] An anomaly detection algorithm is used to detect outliers on the intermediate output features of the lightweight deep neural network model, and the self-supervised training task is adjusted according to the detection results.

[0014] In one embodiment, the step of generating the user's real-time health status using the optimization model includes:

[0015] Based on the dynamic health profile, output a multidimensional health feature vector using the feature extraction layer of the optimization model;

[0016] Inputting the multidimensional health feature vector into the fully connected layer of the optimization model to calculate the health status index;

[0017] The outlier detection result of the multi-dimensional health feature vector is combined with an anomaly detection algorithm to correct the health status indicator and generate a real-time health status.

[0018] In one embodiment, the step of generating an adaptive exercise guidance strategy by using a reinforcement learning algorithm in combination with the user's real-time health status and health trend prediction data includes:

[0019] Constructing a multidimensional state space based on the real-time health status, health trend prediction data, and user feedback data;

[0020] generating initial motion parameters based on the multidimensional state space and detecting real-time health status changes of the user;

[0021] The initial exercise parameters are dynamically adjusted according to the abnormal detection result in the real-time health status, and when the abnormal detection result meets a preset condition, an exercise intensity degradation strategy is triggered to update the exercise guidance strategy.

[0022] In one embodiment, the step of combining the user's real-time health status and health trend prediction data to generate an adaptive exercise guidance strategy through a reinforcement learning algorithm includes:

[0023] Uploading large-scale user data to a cloud server, and using the cloud server to perform deep learning based on the large-scale user data to generate a personalized monitoring algorithm;

[0024] Receive the personalized monitoring algorithm sent by the cloud server, and update the health detection parameters of the lightweight deep neural network model of the terminal device based on the personalized monitoring algorithm.

[0025] In one embodiment, before the step of combining the user's real-time health status and health trend prediction data to generate an adaptive exercise guidance strategy through a reinforcement learning algorithm, the step further includes:

[0026] Adopting a federated learning mechanism, receiving the initial model sent by the cloud server, performing local training on the initial model, and generating a local model gradient;

[0027] Uploading the local model gradient to the cloud server;

[0028] Receiving a first optimization model sent by the cloud server, where the first optimization model is generated by aggregating the local model gradients and updating the global parameters of the initial model based on the cloud server;

[0029] The local model is updated using the first optimized model.

[0030] In one embodiment, applied to a cloud server, the exercise guidance method based on health detection includes:

[0031] Collecting the user's physiological data, behavioral data and environmental data through the terminal device, and pre-processing the physiological data, behavioral data and environmental data to generate the user's dynamic health profile;

[0032] Inputting the dynamic health record into a lightweight deep neural network model through a terminal device, optimizing the lightweight deep neural network model based on the dynamic health record using a self-supervised learning and anomaly detection algorithm, and generating the user's real-time health status using the optimized model;

[0033] Receiving historical health data uploaded by a terminal device, and generating health trend prediction data based on the historical health data;

[0034] Through the terminal device, combined with the user's real-time health status and health trend prediction data, an adaptive exercise guidance strategy is generated through a reinforcement learning algorithm.

[0035] In one embodiment, the step of generating an adaptive exercise guidance strategy by a reinforcement learning algorithm using a terminal device in combination with the user's real-time health status and health trend prediction data includes:

[0036] Receive large-scale user data uploaded by the terminal device, perform deep learning based on the large-scale user data, generate a personalized monitoring algorithm, and send it to the terminal device;

[0037] Through the terminal device, health detection parameters of the lightweight deep neural network model of the terminal device are updated based on the personalized monitoring algorithm.

[0038] In one embodiment, before the step of generating an adaptive exercise guidance strategy through a reinforcement learning algorithm using a terminal device and combining the user's real-time health status and health trend prediction data, the step further includes:

[0039] Adopting a federated learning mechanism, the initial model is sent to each terminal device for model training to generate local model gradients.

[0040] Receiving the local model gradients uploaded by each terminal device, aggregating the local model gradients and updating the global parameters of the initial model to obtain a first optimized model;

[0041] The first optimized model is sent to each terminal device to replace the local old model of each device terminal.

[0042] In addition, to achieve the above-mentioned purpose, the present application further provides a sports guidance device based on health detection, the sports guidance device based on health detection comprising:

[0043] A preprocessing module is used to collect the user's physiological data, behavioral data and environmental data, and preprocess the physiological data, behavioral data and environmental data to generate a dynamic health profile of the user;

[0044] a model optimization module, configured to input the dynamic health profile into a lightweight deep neural network model, optimize the lightweight deep neural network model based on the dynamic health profile using a self-supervised learning and anomaly detection algorithm, and generate the user's real-time health status using the optimized model;

[0045] A prediction module, configured to obtain health trend prediction data issued by a cloud server, wherein the health trend prediction data is generated based on the user's historical health data uploaded by the terminal device;

[0046] The generation module is used to combine the user's real-time health status and health trend prediction data to generate an adaptive exercise guidance strategy through a reinforcement learning algorithm.

[0047] In addition, to achieve the above-mentioned purpose, the present application also provides a terminal device, which includes a memory, a processor, and a health detection-based exercise guidance program stored on the memory and runnable on the processor. When the health detection-based exercise guidance program is executed by the processor, the steps of the health detection-based exercise guidance method as described above are implemented.

[0048] In addition, to achieve the above-mentioned purpose, the present application also provides a computer-readable storage medium, on which a health detection-based exercise guidance program is stored. When the health detection-based exercise guidance program is executed by the processor, the steps of the health detection-based exercise guidance method as described above are implemented.

[0049] One or more technical solutions proposed in this application have at least the following technical effects:

[0050] The technical solution proposed in this application improves the personalized capabilities of health assessment and exercise guidance through the collaborative optimization of multi-dimensional data fusion and intelligent algorithms. First, by collecting physiological data, behavioral data and environmental data, and building a dynamic health profile, a comprehensive digital representation of the user's health status is achieved, providing a data basis for personalized analysis. Secondly, the lightweight deep neural network model of the terminal device dynamically optimizes parameters through self-supervised learning, and combines the anomaly detection algorithm to correct the output in real time, so that the health status assessment can adapt to individual differences, improving the monitoring accuracy and real-time performance. The cloud server performs health trend prediction based on deep learning of historical data, which makes up for the limitations of terminal device computing and realizes long-term risk warning. Finally, the reinforcement learning algorithm dynamically generates exercise guidance strategies through the joint analysis of real-time health status and trend prediction data to ensure that the exercise plan always matches the actual needs of the user. Improve the personalized capabilities of health assessment and exercise guidance for users. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 This is a flowchart of a first exemplary embodiment of the exercise guidance method based on health detection of the present application;

[0052] Figure 2 This is a flowchart of a second exemplary embodiment of the exercise guidance method based on health detection of the present application;

[0053] Figure 3 This is a flowchart of a third exemplary embodiment of the exercise guidance method based on health detection of the present application;

[0054] Figure 4 This is a flowchart of a fourth exemplary embodiment of the exercise guidance method based on health detection of the present application;

[0055] Figure 5 This is a schematic diagram of the module structure of the exercise guidance device based on health detection in an embodiment of the present application;

[0056] Figure 6 This is a schematic diagram of the device structure of the hardware operating environment involved in the exercise guidance method based on health detection in an embodiment of the present application.

[0057] The realization of the objectives, functional features and advantages of this application will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0058] It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0059] The main technical solution of this application is: collecting the user's physiological data, behavioral data and environmental data through terminal devices, and preprocessing the physiological data, behavioral data and environmental data to generate the user's dynamic health record; inputting the dynamic health record into a lightweight deep neural network model, using self-supervised learning and anomaly detection algorithms to optimize the lightweight deep neural network model based on the dynamic health record, and using the optimization model to generate the user's real-time health status; receiving historical health data uploaded by the terminal device through a cloud server, and generating health trend prediction data based on the historical health data; combining the user's real-time health status and health trend prediction data through the terminal device, and generating an adaptive exercise guidance strategy through a reinforcement learning algorithm.

[0060] This application actually takes into account the lack of personalization in existing smart wearable devices in terms of health monitoring and exercise guidance. Traditional technologies rely on fixed algorithms and preset models, and cannot dynamically adjust to the individual physiological differences of users, resulting in limited accuracy of health assessments. In addition, existing solutions lack the ability to deeply analyze long-term health data, making it difficult to predict potential risks, and exercise guidance strategies often use universal templates and cannot provide accurate suggestions based on the user's real-time status. These problems restrict the scientific nature of health management and user experience.

[0061] Based on this, the present application proposes a method and device for exercising guidance based on health detection, which is applied to terminal devices.

[0062] Specifically, the following are the detailed steps of the first exemplary embodiment of the exercise guidance method based on health detection in this application:

[0063] Reference Figure 1 , Figure 1 This is a flowchart of a first exemplary embodiment of the exercise guidance method based on health detection in this application. In this embodiment, the exercise guidance method based on health detection includes steps S10 to S40:

[0064] Step S10, collecting the user's physiological data, behavioral data and environmental data, and pre-processing the physiological data, behavioral data and environmental data to generate a dynamic health profile of the user;

[0065] Regarding data collection, one feasible implementation involves using a PPG sensor to collect real-time monitoring of a user's heart rate and blood oxygen saturation, an ECG sensor to acquire electrocardiogram signals, and a temperature sensor to record physiological data such as changes in body surface temperature. Accelerometers and gyroscopes are used to capture motion characteristics such as the user's movement trajectory, cadence, and posture, and combined with voice data collected by a microphone to assist in determining behavioral data such as the user's activity status. Temperature and humidity sensors and a barometer are used to record environmental data such as the temperature, humidity, and altitude changes of the user's environment.

[0066] Regarding data preprocessing, one feasible implementation involves first filtering and denoising the collected raw data. For example, a Kalman filter is used to remove motion artifacts from the PPG signal, and a sliding window algorithm is used to smooth the accelerometer data to eliminate transient jitter. Subsequently, a feature extraction module is used to align and fuse the multimodal data in time and space. For example, heart rate variability is correlated with the time series of exercise intensity, and temperature readings are corrected using ambient temperature and humidity data. The resulting dynamic health record is stored in a structured format and serves as a unified input for subsequent analysis, ensuring data consistency and traceability.

[0067] Step S20: inputting the dynamic health record into a lightweight deep neural network model, optimizing the lightweight deep neural network model based on the dynamic health record using a self-supervised learning and anomaly detection algorithm, and generating the user's real-time health status using the optimized model;

[0068] In a feasible implementation, the step of optimizing the lightweight deep neural network model based on the dynamic health record using self-supervised learning and anomaly detection algorithms includes steps A1 to A3:

[0069] Step A1, constructing a self-supervised training task based on the dynamic health profile;

[0070] Specifically, the multimodal time series data in the dynamic health records is first preprocessed. A sliding window technique is used to segment the continuous physiological, behavioral, and environmental data into time series samples in fixed time windows, forming structured input data. Three typical approaches can be used to construct self-supervised training tasks based on these time series samples: First, the sensor data reconstruction task randomly masks 15% to 30% of the feature dimensions in the input data, requiring the model to predict the masked feature values ​​based on the unmasked contextual information; second, the future time step prediction task, which uses data from the previous N time steps as input, such as the first 10 time windows, and trains the model to predict changes in physiological indicators for the next M time steps; and third, the abnormal sample comparative learning task, which extracts normal samples and known abnormal samples from historical data and uses comparative learning to allow the model to learn the characteristic differences between normal and abnormal data.

[0071] Step A2: updating the health detection parameters of the lightweight deep neural network model using the loss function of the self-supervised training task;

[0072] Specifically, for the self-supervised training task constructed above, a lightweight deep neural network model with an encoder-decoder architecture is used for training. Taking the data reconstruction task as an example, the encoder maps the input masked health data into a low-dimensional feature representation, and the decoder reconstructs the complete original data based on this feature representation. The loss function uses mean squared error to calculate the difference between the reconstructed data and the original data, focusing on optimizing the accuracy of the restoration of key physiological indicators. Through the backpropagation algorithm, this loss function is used to iteratively update the model's health detection parameters, gradually improving the model's ability to capture individual health characteristics of users.

[0073] Step A3: Use an anomaly detection algorithm to perform outlier detection on the intermediate output features of the lightweight deep neural network model, and adjust the self-supervised training task according to the detection results.

[0074] Specifically, during the model training process, the multi-dimensional health feature vectors output by the encoder's intermediate layer are extracted, and outlier detection algorithms (such as isolation forest or K-Means clustering) are used to detect these feature vectors. If the proportion of outliers detected exceeds the preset threshold, it means that the current self-supervised training task fails to effectively cover the health data characteristics of the individual user, and the task construction method needs to be adjusted: for example, if the outliers in the data reconstruction task are concentrated on the heart rate characteristics, the mask ratio of the heart rate dimension can be increased or the randomness of the mask position can be adjusted; if the outliers in the future time step prediction task are related to sudden changes in motion state, the time window length is expanded to include more contextual information. By dynamically adjusting the self-supervised training task, it is ensured that the model can continuously adapt to the health data distribution of the individual user, and ultimately optimize the personalized lightweight deep neural network model for the user.

[0075] In a feasible implementation, the step of generating the user's real-time health status using the optimization model includes steps B1 to B3:

[0076] Step B1, based on the dynamic health record, using the feature extraction layer of the optimization model to output a multi-dimensional health feature vector;

[0077] Specifically, after the optimization of the lightweight deep neural network model is completed, a dynamic health profile is generated based on the feature extraction layer of the optimized model. The dynamic health profile contains multimodal data that has been filtered, denoised, and aligned in time and space, including physiological data and behavioral data. The feature extraction layer, as the core encoding module of the model, is usually composed of a convolutional layer or a recurrent neural network layer. It traverses the time series data through a sliding window and performs feature fusion and abstraction on the data of each dimension: for example, the convolutional layer captures the synchronous change characteristics of heart rate and movement acceleration in a short period of time through the local receptive field, and the recurrent layer extracts the long-term correlation pattern of body temperature and ambient humidity through the memory unit. Finally, the feature extraction layer outputs a health feature vector containing multi-dimensional information. This vector comprehensively reflects the composite characteristics of the user's current physiological state, behavioral pattern, and environmental influences, such as "blood oxygen-heart rate coupling characteristics during exercise", "body temperature-ambient humidity correlation characteristics during sleep", etc.

[0078] Step B2, inputting the multidimensional health feature vector into the fully connected layer of the optimization model to calculate the health status index;

[0079] Specifically, the multidimensional health feature vector output from step B1 is input into the fully connected layer of the model. The fully connected layer serves as a feature decoding module, with its neuron nodes corresponding to evaluation indicators for different health dimensions. The feature vectors are linearly combined using trained weight coefficients between nodes, and nonlinearly mapped using an activation function to ultimately calculate a quantitative health status indicator. For example, if the "blood oxygen-heart rate coupling feature during exercise" in the feature vector has a high and stable value, the fully connected layer will assign it a higher cardiopulmonary function index weight based on the training parameters, thereby outputting an indicator value reflecting the user's current good cardiopulmonary function.

[0080] Step B3: combining an outlier detection result of the multi-dimensional health feature vector with an anomaly detection algorithm, correcting the health status indicator, and generating a real-time health status.

[0081] Specifically, while extracting the multidimensional health feature vector in step B1, the trained anomaly detection algorithm is synchronously called to perform outlier detection on the multidimensional health feature vector. If it is detected that the characteristics of a certain dimension in the feature vector significantly deviate from the normal range of the user's historical data distribution, it is determined that there is an outlier in this dimension. At this time, the health status index calculated in step B2 is corrected based on the severity of the outlier: if it is a mild outlier, the indicator value of the corresponding dimension is smoothed; if it is a severe outlier, the overall health status index is directly lowered and marked as "potential abnormality". Through this correction mechanism, the real-time health status finally generated can more accurately reflect the user's current true health level and avoid misjudgment of indicators due to abnormal characteristics of a single dimension.

[0082] Step S30, obtaining health trend prediction data issued by a cloud server, wherein the health trend prediction data is generated based on the historical health data of the user uploaded by the terminal device;

[0083] A feasible implementation method is applied to a cloud server to receive historical health data uploaded by a terminal device and generate health trend prediction data based on the historical health data;

[0084] Specifically, the terminal device regularly uploads the user's historical health data, including pre-processed physiological, behavioral, and environmental data stored in the dynamic health profile, to a cloud server via a secure communication protocol. The cloud server first cleans and standardizes the received historical data by filling in missing values, filtering out noise, and aligning multimodal data by timestamp to form an input dataset in a unified time series format.

[0085] The cloud server then uses a pre-trained deep learning model (such as an LSTM long short-term memory network or a Transformer model) to predict health trends. Taking the LSTM model as an example, it uses the memory units of a recurrent neural network to capture long-term dependencies in historical data. For example, it analyzes the correlation patterns between a user's exercise heart rate, blood oxygen saturation, and ambient temperature over the past six months to identify the underlying pattern of "an increase in resting heart rate the day after exercise in a high-temperature environment." The model input is standardized multidimensional time series data, such as a one-day time step, with each step containing 10 to 15 features such as heart rate mean, maximum exercise intensity, sleep efficiency, and ambient humidity. The output is the predicted value of health indicators for the next N days, such as the resting heart rate fluctuation range, exercise endurance change trend, and potential dehydration risk probability.

[0086] For personalized trend predictions, the cloud server further fine-tunes the global model based on the user's individual characteristics. For example, for users with a history of hypertension, the model will increase the weight of blood pressure-related physiological indicators to improve the accuracy of cardiovascular risk prediction. After the prediction is completed, the cloud server verifies the fit between the prediction result and the historical data through residual analysis. If the error exceeds the threshold, the model self-calibration process is triggered to adjust the network parameters to optimize the subsequent prediction accuracy. Finally, the cloud server encapsulates the generated health trend prediction data into structured data and sends it to the terminal device. Through this process, the cloud server effectively compensates for the limitations of the terminal device's computing resources, uses large-scale data and complex models to achieve long-term dynamic prediction of the user's health status, and provides a trend basis for personalized exercise guidance.

[0087] Step S40: generating an adaptive exercise guidance strategy through a reinforcement learning algorithm based on the user's real-time health status and health trend prediction data.

[0088] In a feasible implementation, step S40 may include steps S41 to S43:

[0089] Step S41, constructing a multidimensional state space based on the real-time health status, health trend prediction data and user feedback data;

[0090] Specifically, the system first integrates the user's real-time health status, cloud-generated health trend prediction data, and historical user feedback data to construct the multidimensional state space required for reinforcement learning. Real-time health status includes quantitative indicators after feature extraction and anomaly correction, such as a cardiopulmonary function index of 82 points, metabolic activity of 75 points, and exercise adaptability of 68 points. Health trend prediction data includes potential changes over the next seven days, such as "Post-exercise recovery time is expected to be extended by 12%" and "Resting heart rate fluctuation range expands to 5-10 beats / minute." User feedback data includes subjective evaluations of historical exercise plans, such as "Strong muscle soreness after interval training last week" and "High acceptance of endurance training." These multi-source data are standardized and aligned by timestamp, ultimately forming a state vector containing dimensions such as physiological state, trend risk, and user preference, which serves as the input state space of the reinforcement learning algorithm. For example, the state vector [0.82, 0.75, 0.68, 0.12, -0.05, 0.7] corresponds to the cardiopulmonary index, metabolic activity, exercise adaptability, recovery time extension ratio, resting heart rate fluctuation offset, and endurance training preference, respectively.

[0091] Step S42, generating initial motion parameters based on the multidimensional state space, and detecting real-time health status changes of the user;

[0092] Specifically, based on the multidimensional state space, the reinforcement learning algorithm generates initial motion parameters through the policy network. The output of the policy network corresponds to the core parameters of the exercise plan, including exercise type, duration, interval arrangement, etc. When generating parameters, the algorithm gives priority to matching user preferences, while combining real-time health status and trend prediction. After the initial parameters are generated, the terminal device collects the user's physiological data during exercise in real time, and compares it with the expected range of the initial parameters to monitor changes in health status: for example, if the user's heart rate continues to be higher than 145 beats / minute after 5 minutes of exercise, or the blood oxygen saturation is lower than 95%, it will be marked as "abnormal state fluctuation".

[0093] Step S43: dynamically adjusting the initial exercise parameters according to the abnormal detection result in the real-time health status, and triggering the exercise intensity degradation strategy to update the exercise guidance strategy when the abnormal detection result meets the preset conditions.

[0094] Specifically, the initial motion parameters are dynamically adjusted based on the abnormal results of the real-time health status detected. The abnormality detection results include the type, severity and duration of the outlier. If a mild abnormality is detected, the motion parameters are fine-tuned; if a severe abnormality is detected, the exercise intensity degradation strategy is triggered: the current exercise type is switched from running to low-intensity cycling, the target intensity is lowered to 50% to 60% of the maximum heart rate, and the exercise is forced to pause for 1 minute for recovery. The adjusted parameters are fed back to the user through the terminal device screen or voice prompts, and the reinforcement learning reward function is updated at the same time, ultimately forming a dynamic exercise guidance strategy that adapts to the user's current state.

[0095] Further, when applied to a terminal device, refer to Figure 2 , Figure 2 This is a flow chart of a second exemplary embodiment of the exercise guidance method based on health detection in this application. In this embodiment, the step S40 may include steps C1 to C2:

[0096] Step C1: uploading large-scale user data to a cloud server, and generating a personalized monitoring algorithm by performing deep learning based on the large-scale user data by the cloud server;

[0097] Step C2: Receive the personalized monitoring algorithm sent by the cloud server, and update the health detection parameters of the lightweight deep neural network model of the terminal device based on the personalized monitoring algorithm.

[0098] When applied to a cloud server, this embodiment includes steps C3 to C4:

[0099] Step C3: receiving large-scale user data uploaded by the terminal device, performing deep learning based on the large-scale user data, generating a personalized monitoring algorithm, and sending the algorithm to the terminal device;

[0100] Step C4: updating the health detection parameters of the lightweight deep neural network model of the terminal device based on the personalized monitoring algorithm through the terminal device.

[0101] In one feasible implementation, the terminal device regularly uploads the dynamic health records of a large number of users to the cloud server through a secure communication protocol. The cloud server first pre-processes the received raw data in a unified manner to form a standardized user health data set. Based on the pre-processed large-scale data, the cloud server calls the deep learning model to mine group health patterns and generate personalized algorithms. Taking the LSTM model as an example, it captures the long-term health feature associations of different user groups through the memory units of the recurrent neural network: for example, it analyzes the correlation pattern between the exercise heart rate, sleep efficiency and ambient temperature of female users aged 20-30, and identifies the common law of "decreased sleep quality after high-intensity exercise in a low-temperature environment"; at the same time, for the historical data of individual users, the model strengthens the weight allocation of individual features through the attention mechanism.

[0102] After model training, the cloud server generates a personalized monitoring algorithm based on the user's historical data and group pattern analysis results. This algorithm specifically implements rules for adjusting the parameters of the terminal device's lightweight deep neural network model. For example, if user A's historical data shows that their heart rate recovery time after exercise is 20% slower than the group average, the algorithm will increase the weight of the "heart rate recovery feature" in the model, such as adjusting the kernel function coefficient of the corresponding convolutional layer from 0.2 to 0.4. If user B's sleep cycle is often affected by ambient humidity, the algorithm will strengthen the fusion parameters of the "humidity-sleep correlation feature" in the model, such as adjusting the bias value of the humidity feature in the fully connected layer from -0.1 to 0.2.

[0103] The generated personalized monitoring algorithm is sent to the terminal device. After receiving the parameter package, the terminal device performs a targeted update on the health monitoring parameters of the local lightweight deep neural network model. For example, according to the algorithm instructions, the convolution kernel weight corresponding to the "heart rate recovery feature" is increased by 0.2, or the bias value of the fully connected layer of the "humidity-sleep correlation feature" is adjusted. After the update is completed, the terminal device verifies the consistency of the model output using local test data to ensure that the personalized monitoring algorithm effectively improves the model's adaptability to the user's individual health data.

[0104] Based on this, the cloud server uses the deep learning capabilities of large-scale user data to make up for the limitations of terminal devices relying solely on individual data training, which not only ensures the universality of the model but also enhances personalization, ultimately achieving continuous improvement in the accuracy of terminal device health monitoring.

[0105] Further, when applied to a terminal device, refer to Figure 3 , Figure 3 This is a flowchart of the third exemplary embodiment of the exercise guidance method based on health detection in this application. In this embodiment, the step S40 may further include steps D1 to D4:

[0106] Step D1: using a federated learning mechanism, receiving the initial model sent by the cloud server, performing local training on the initial model, and generating a local model gradient;

[0107] Step D2, uploading the local model gradient to the cloud server;

[0108] Step D3, receiving a first optimization model sent by the cloud server, wherein the first optimization model is generated by aggregating the local model gradients and updating the global parameters of the initial model based on the cloud server;

[0109] Step D4: using the first optimization model to update the local model.

[0110] When applied to a cloud server, this embodiment includes steps D5 to D7:

[0111] Step D5: Using the federated learning mechanism, the initial model is sent to each terminal device for model training to generate local model gradients.

[0112] Step D6: receiving the local model gradients uploaded by each terminal device, aggregating the local model gradients and updating the global parameters of the initial model to obtain a first optimized model;

[0113] Step D7: Send the first optimized model to each terminal device to replace the local old model of each device terminal.

[0114] In one feasible implementation, a cloud server first sends the initial model to each user's smart wearable terminal device. After receiving the model, the terminal device performs local training based on the user's locally stored dynamic health profile. During training, the terminal device only uses the user's private data and does not transmit raw data to protect privacy. Training uses stochastic gradient descent or the Adam optimizer, using the user's real-time health status prediction error as the loss function. The gradient of the model parameters is calculated through backpropagation, ultimately generating a data packet containing the local model gradient information, which only contains the gradient value and no raw data.

[0115] Each terminal device uploads its locally generated gradient data packet to the cloud server via an encrypted protocol. After receiving the gradient data from all devices, the cloud server performs gradient aggregation: first, a weighted coefficient is assigned based on the amount of local data on each device. Then, the gradients are weighted averaged using a federated averaging algorithm to obtain a global gradient update. For example, if the gradient of device A is [0.02, -0.01], representing 30% of the data volume, and the gradient of device B is [0.01, 0.03], representing 70% of the data volume, the aggregated global gradient is [0.02 × 0.3 + 0.01 × 0.7, -0.01 × 0.3 + 0.03 × 0.7] = [0.013, 0.018]. Based on this, the cloud server updates the global parameters of the initial model, forming a first optimization model that integrates the data characteristics of multiple devices, preserving the characteristics of individual data while preventing leakage of the original data.

[0116] The cloud server encapsulates the parameters of the first optimization model into a parameter package and sends it to each terminal device. After receiving the parameter package, the terminal device first verifies the parameter integrity and then replaces the corresponding parameters of the local old model with the new parameters. For example, the original feature extraction layer weights [0.1, 0.2] are replaced with [0.1 + 0.013, 0.2 + 0.018] = [0.113, 0.218]. After the replacement is complete, the terminal device verifies the new model using local recent health data: the mean squared error between the health status indicators output by the model and the actual measurements is calculated. If the error is below a preset threshold, the model is confirmed to be valid and the first optimization model is officially activated. If the error is too large, the old model is rolled back and the anomaly is reported to the cloud, triggering re-aggregation or adjusting the aggregation strategy. Through this process, the federated learning mechanism protects user privacy while leveraging multi-device data to collaboratively optimize the model, significantly improving the generalization and personalized adaptation of the terminal device health monitoring model.

[0117] Further, refer to Figure 4 , Figure 4 This is a flowchart of a fourth exemplary embodiment of the exercise guidance method based on health detection in this application. In this embodiment, based on the first embodiment, the step S40 may include steps S51 to S52:

[0118] Step S51, monitoring the user's real-time physiological data. If it exceeds a preset safety threshold, triggering a level 1 alarm through tactile feedback and voice prompts, and forcibly pausing the current exercise plan and switching to a safe mode;

[0119] Specifically, the terminal device collects the user's physiological data in real time through built-in sensors, and transmits the data to the terminal device at a fixed frequency for health detection, and sets personalized safety thresholds based on the user's dynamic health profile: for example, when the heart rate exceeds 100 beats / minute in a resting state or the blood oxygen level is lower than 95% during exercise, a level 1 alarm condition is triggered.

[0120] When real-time physiological data exceeds a preset threshold, the device immediately triggers a Level 1 alarm through tactile feedback (e.g., three consecutive short vibrations at 100Hz) and voice prompts (e.g., "Abnormal blood oxygen levels detected, exercise paused"). Simultaneously, the device forcibly interrupts the current exercise plan and switches to safe mode. In safe mode, the device reduces sensor sampling frequency to reduce power consumption, retaining only the monitoring function for core physiological indicators, and continuously displays the prompt "Safe Mode, Rest Recommended" on the screen until the user actively confirms or the physiological data returns to within the safe threshold.

[0121] Step S52: When the cloud server identifies that the user's historical health data has a predefined combined risk pattern, it generates a secondary warning report containing medical intervention suggestions and pushes it to the user through the terminal device.

[0122] Specifically, the cloud server regularly conducts in-depth analysis of users' historical health data and performs pattern matching based on predefined risk combination patterns. These combination patterns are jointly developed by medical experts and data scientists, combining clinical guidelines with statistical patterns of large-scale user data.

[0123] When the cloud identifies a user's historical data matching a specific risk pattern, the server leverages the medical knowledge graph to generate specific medical intervention recommendations. This generated secondary warning report, containing a risk description, specific data supporting the recommendations, and medical advice, is delivered to the user via Bluetooth or network on the device. The device then prompts the user to review the report through an app pop-up, scrolling display on the device screen, or voice notification, ensuring timely access to professional health guidance and mitigation of potential health risks.

[0124] In addition, this application also proposes a sports guidance device based on health detection, such as Figure 5 As shown, the exercise guidance device based on health detection includes:

[0125] The preprocessing module 10 is used to collect the user's physiological data, behavioral data and environmental data, and preprocess the physiological data, behavioral data and environmental data to generate the user's dynamic health profile;

[0126] a model optimization module 20 for inputting the dynamic health profile into a lightweight deep neural network model, optimizing the lightweight deep neural network model based on the dynamic health profile using a self-supervised learning and anomaly detection algorithm, and generating the user's real-time health status using the optimized model;

[0127] A prediction module 30 is configured to obtain health trend prediction data issued by a cloud server, wherein the health trend prediction data is generated based on the user's historical health data uploaded by the terminal device;

[0128] The generation module 40 is used to generate an adaptive exercise guidance strategy through a reinforcement learning algorithm in combination with the user's real-time health status and health trend prediction data.

[0129] The health-detection-based exercise guidance device provided in this application adopts the health-detection-based exercise guidance method in the above-mentioned embodiment, aiming to improve the personalized ability to perform health assessments and exercise guidance on users. Compared with the prior art, the beneficial effects of the health-detection-based exercise guidance device provided in this application are the same as those of the health-detection-based exercise guidance method in the above-mentioned embodiment, and the other technical features of the health-detection-based exercise guidance device are the same as those disclosed in the above-mentioned embodiment method, and are not further described here.

[0130] The present application provides an exercise guidance device based on health detection, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the exercise guidance method based on health detection in the above-mentioned embodiment one.

[0131] The health detection-based exercise guidance equipment in the embodiments of the present application may include but is not limited to mobile terminals such as mobile phones, laptop computers, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Descriptions), PMPs (Portable Media Players), vehicle-mounted terminals (such as vehicle-mounted navigation terminals), etc., as well as fixed terminals such as digital TVs, desktop computers, etc. Figure 6 The health detection-based exercise guidance device shown is merely an example and should not limit the functions and scope of use of the embodiments of the present application.

[0132] like Figure 6As shown, the health-based exercise guidance device may include a processing device 1001 (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes based on programs stored in a read-only memory 1002 or programs loaded from a storage device 1003 into a random access memory 1004. The random access memory 1004 also stores various programs and data required for the operation of the health-based exercise guidance device. The processing device 1001, the read-only memory 1002, and the random access memory 1004 are interconnected via a bus 1005. An input / output interface 1006 is also connected to the bus. Typically, the following systems can be connected to the input / output interface 1006: an input device 1007 including, for example, a touch screen, touchpad, keyboard, mouse, image sensor, microphone, accelerometer, gyroscope, etc.; an output device 1008 including, for example, a liquid crystal display (LCD), speaker, vibrator, etc.; a storage device 1003 including, for example, a magnetic tape, hard disk, etc.; and a communication device 1009. The communication device 1009 can allow the health detection-based sports coaching device to communicate with other devices wirelessly or by wire to exchange data. Although the figure shows a health detection-based sports coaching device with various systems, it should be understood that it is not required to implement or have all of the systems shown. More or fewer systems may be implemented or have alternatively.

[0133] In particular, according to the embodiments disclosed in the present application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, the embodiments disclosed in the present application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program comprising program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network via a communication device, or installed from a storage device 1003, or installed from a read-only memory 1002. When the computer program is executed by the processing device 1001, the above-mentioned functions defined in the method of the embodiment disclosed in the present application are performed.

[0134] The health detection-based exercise guidance device provided in this application adopts the health detection-based exercise guidance method in the above-mentioned embodiment, aiming to improve the personalized ability to perform health assessments and exercise guidance on users. Compared with the prior art, the beneficial effects of the health detection-based exercise guidance device provided in this application are the same as the beneficial effects of the health detection-based exercise guidance method provided in the above-mentioned embodiment, and the other technical features of the health detection-based exercise guidance device are the same as those disclosed in the method of the previous embodiment, and are not further described here.

[0135] It should be understood that the various parts disclosed in this application can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any one or more embodiments or examples in a suitable manner.

[0136] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

[0137] The present application provides a computer-readable storage medium having computer-readable program instructions (ie, a computer program) stored thereon, and the computer-readable program instructions are used to execute the exercise guidance method based on health detection in the above-mentioned embodiment.

[0138] The computer-readable storage medium provided in this application may be, for example, a USB flash drive, but is not limited to electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems or devices, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In this embodiment, the computer-readable storage medium may be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system or device. The program code contained on the computer-readable storage medium may be transmitted using any appropriate medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination thereof.

[0139] The computer-readable storage medium may be included in the exercise guidance device based on health detection, or may exist independently without being assembled into the exercise guidance device based on health detection.

[0140] Computer program code for performing the operations of the present application may be written in one or more programming languages, or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a stand-alone software package, partially on the user's computer and partially on a remote computer, or entirely on the remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., through the Internet using an Internet service provider).

[0141] The flowcharts and block diagrams in the accompanying drawings illustrate the possible architecture, functions and operations of the systems, methods and computer program products according to various embodiments of the present application. In this regard, each box in the flowchart or block diagram can represent a module, program segment or a part of code, and the module, program segment or a part of code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order than that marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, and the combination of the boxes in the block diagram and / or flowchart, can be implemented by a dedicated hardware-based system that performs the specified function or operation, or can be implemented by a combination of dedicated hardware and computer instructions.

[0142] The modules described in the embodiments of the present application may be implemented in software or hardware, wherein the name of a module does not necessarily limit the unit itself.

[0143] The readable storage medium provided in this application is a computer-readable storage medium, which stores computer-readable program instructions (i.e., a computer program) for executing the above-mentioned health detection-based exercise guidance method, and is intended to improve the personalized ability to perform health assessments and exercise guidance on users. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided in this application are the same as the beneficial effects of the health detection-based exercise guidance method provided in the above-mentioned embodiment, and will not be repeated here.

[0144] The present application also provides a computer program product, comprising a computer program, which implements the steps of the above-mentioned health detection-based exercise guidance method when executed by a processor.

[0145] The computer program product provided in this application is intended to enhance the ability to provide personalized health assessments and exercise guidance to users. Compared to the prior art, the beneficial effects of the computer program product provided in this application are the same as those of the health assessment-based exercise guidance method provided in the above-mentioned embodiments, and are not further elaborated here.

[0146] Compared with the prior art, the exercise guidance method, device, equipment, medium and computer product based on health detection proposed in the embodiment of the present application extracts the business characteristic information of the target business, performs data standardization processing on the business characteristic information to obtain standard characteristic data, performs hash processing on the standard characteristic data to obtain unique characteristic data, performs numerical processing and splicing processing on the unique characteristic data to obtain the first business characteristic value, accumulates the first business characteristic value of the target business to obtain the target business characteristic value, and finally compares the target business characteristic value with the characteristic value set to obtain the exercise guidance result based on health detection. Compared with the traditional method of identifying repeated businesses by generating a unique key value or a continuous serial number for each business, it is more efficient, flexible and reliable. Based on the solution of the present application, the business of complex scenarios is transformed into a comparison of two numbers through a series of simple transformations, making the comparison process very intuitive and efficient. The system only needs to simply compare whether the two values ​​are equal to quickly determine whether the two businesses are exactly the same.

[0147] It should be noted that, in this document, the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, article, or system comprising a series of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or system. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, method, article, or system comprising the element.

[0148] The serial numbers of the above embodiments of the present application are for description only and do not represent the advantages or disadvantages of the embodiments.

[0149] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus the necessary general hardware platform, and of course can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as mentioned above, and includes a number of instructions for enabling a terminal device (which can be a mobile phone, computer, server, controlled terminal, or network device, etc.) to execute the method of each embodiment of the present application.

[0150] The above are only preferred embodiments of the present application and do not limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made using the contents of the present application specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present application.

Claims

1. A method of exercise guidance based on health detection, characterized in that: Applied to a terminal device, the exercise guidance method based on health detection includes: Collecting the user's physiological data, behavioral data and environmental data, and pre-processing the physiological data, behavioral data and environmental data to generate a dynamic health profile of the user; Inputting the dynamic health record into a lightweight deep neural network model, optimizing the lightweight deep neural network model based on the dynamic health record using a self-supervised learning and anomaly detection algorithm, and generating the user's real-time health status using the optimized model; Obtaining health trend prediction data issued by a cloud server, where the health trend prediction data is generated based on the user's historical health data uploaded by the terminal device; Combining the user's real-time health status and health trend prediction data, an adaptive exercise guidance strategy is generated through a reinforcement learning algorithm.

2. The exercise guidance method based on health detection according to claim 1, characterized in that: The step of optimizing the lightweight deep neural network model based on the dynamic health record using self-supervised learning and anomaly detection algorithms includes: Based on the dynamic health profile, construct a self-supervised training task; Updating health detection parameters of the lightweight deep neural network model using the loss function of the self-supervised training task; An anomaly detection algorithm is used to detect outliers on the intermediate output features of the lightweight deep neural network model, and the self-supervised training task is adjusted according to the detection results.

3. The exercise guidance method based on health detection according to claim 2, characterized in that: The step of generating the user's real-time health status by using the optimization model comprises: Based on the dynamic health profile, output a multidimensional health feature vector using the feature extraction layer of the optimization model; Inputting the multidimensional health feature vector into the fully connected layer of the optimization model to calculate the health status index; The outlier detection result of the multi-dimensional health feature vector is combined with an anomaly detection algorithm to correct the health status indicator and generate a real-time health status.

4. The exercise guidance method based on health detection according to claim 1, characterized in that: The step of generating an adaptive exercise guidance strategy by using a reinforcement learning algorithm in combination with the user's real-time health status and health trend prediction data includes: Constructing a multidimensional state space based on the real-time health status, health trend prediction data, and user feedback data; generating initial motion parameters based on the multidimensional state space, and detecting real-time health status changes of the user; The initial exercise parameters are dynamically adjusted according to the abnormal detection result in the real-time health status, and when the abnormal detection result meets a preset condition, an exercise intensity degradation strategy is triggered to update the exercise guidance strategy.

5. The exercise guidance method based on health detection according to claim 1, characterized in that: The step of combining the user's real-time health status and health trend prediction data to generate an adaptive exercise guidance strategy through a reinforcement learning algorithm includes: Uploading large-scale user data to a cloud server, and using the cloud server to perform deep learning based on the large-scale user data to generate a personalized monitoring algorithm; Receive the personalized monitoring algorithm sent by the cloud server, and update the health detection parameters of the lightweight deep neural network model of the terminal device based on the personalized monitoring algorithm.

6. The exercise guidance method based on health detection according to claim 1, characterized in that: Before the step of combining the user's real-time health status and health trend prediction data to generate an adaptive exercise guidance strategy through a reinforcement learning algorithm, the following steps are further included: Adopting a federated learning mechanism, receiving the initial model sent by the cloud server, performing local training on the initial model, and generating a local model gradient; Uploading the local model gradient to the cloud server; Receiving a first optimization model sent by the cloud server, where the first optimization model is generated by aggregating the local model gradients and updating the global parameters of the initial model based on the cloud server; The local model is updated using the first optimized model.

7. A method of exercise guidance based on health detection, characterized in that: Applied to a cloud server, the exercise guidance method based on health detection includes: Collecting the user's physiological data, behavioral data and environmental data through the terminal device, and pre-processing the physiological data, behavioral data and environmental data to generate the user's dynamic health profile; Inputting the dynamic health record into a lightweight deep neural network model through a terminal device, optimizing the lightweight deep neural network model based on the dynamic health record using a self-supervised learning and anomaly detection algorithm, and generating the user's real-time health status using the optimized model; Receiving historical health data uploaded by a terminal device, and generating health trend prediction data based on the historical health data; Through the terminal device, combined with the user's real-time health status and health trend prediction data, an adaptive exercise guidance strategy is generated through a reinforcement learning algorithm.

8. The exercise guidance method based on health detection according to claim 7, characterized in that: The step of generating an adaptive exercise guidance strategy by a reinforcement learning algorithm using a terminal device in combination with the user's real-time health status and health trend prediction data includes: Receive large-scale user data uploaded by the terminal device, perform deep learning based on the large-scale user data, generate a personalized monitoring algorithm, and send it to the terminal device; Through the terminal device, health detection parameters of the lightweight deep neural network model of the terminal device are updated based on the personalized monitoring algorithm.

9. The exercise guidance method based on health detection according to claim 7, characterized in that: Before the step of generating an adaptive exercise guidance strategy by a reinforcement learning algorithm using a terminal device and combining the user's real-time health status and health trend prediction data, the step further includes: Adopting a federated learning mechanism, the initial model is sent to each terminal device for model training to generate local model gradients. Receiving the local model gradients uploaded by each terminal device, aggregating the local model gradients and updating the global parameters of the initial model to obtain a first optimized model; The first optimized model is sent to each terminal device to replace the local old model of each device terminal.

10. A sports guidance device based on health detection, characterized in that: The device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program is configured to implement the steps of the exercise guidance method based on health detection according to any one of claims 1 to 9.

Images