Intelligent risk early warning interaction system for ankle pump training and intelligent ankle pump trainer

Abstract

The invention discloses an intelligent risk early warning interaction system for ankle pump training and an intelligent ankle pump trainer, and relates to the technical field of risk early warning. The intelligent risk early warning interaction system for ankle pump training comprises a data acquisition module used for acquiring training data of a plurality of users, and a feature extraction module used for analyzing a behavior control ability feature value and a physiological stress response feature value based on behavior data and physiological feedback data of each user; the guide analysis module is used for analyzing a guide deviation characteristic value based on the virtual target video stream data in combination with a pre-trained guide response alignment recognition model; the risk analysis module is used for analyzing the training risk characteristic values based on the characteristic values, risk interaction early warning is performed on each user through the training risk characteristic values in the early warning interaction module, so that the risk identification precision is remarkably improved, the system performs accurate early warning, and the user experience is improved. Therefore, the accuracy and effectiveness of ankle pump movement and the compliance of ankle joint movement of a patient are ensured.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of risk early warning, in particular to an intelligent risk early warning interaction system for ankle pump training and an intelligent ankle pump trainer. BACKGROUND

[0002] Ankle pump training is a kind of active lower limb exercise commonly used to prevent lower limb venous thrombosis and improve peripheral blood circulation, and is widely used in postoperative rehabilitation, elderly care and rehabilitation management of long-term bedridden patients. Through repeated dorsiflexion and plantar flexion movements, the training method activates the calf muscle pump function, thereby promoting venous blood return and improving lower limb circulation efficiency. However, some users may experience high-risk states such as excessive exercise, loss of control, and excessive stress during training due to insufficient neural control, muscle fatigue, and abnormal blood flow dynamics. However, the existing system generally lacks a dynamic risk identification mechanism, making it difficult to provide early warning prompts or interactive interventions.

[0003] The prior art, such as the patent application with publication number CN115844359B, discloses a risk early warning interaction system for intelligent ankle pump training detection and analysis and an intelligent ankle ring. The system includes: collecting information through an intelligent ankle ring, wherein the collected information includes ankle pump movement information of the wearer of the intelligent ankle ring and / or blood flow rate values of the wearer; the intelligent ankle ring uploads the collected information to the cloud for intelligent analysis; the intelligent ankle ring analyzes the wearer's achievement of the training task locally and sends it to the cloud; the cloud determines whether the wearer has completed the training based on the achievement; and the intelligent ankle ring provides interactive prompt information to the wearer based on the completion of the training completion indicator, wherein the interactive prompt information includes encouragement prompt information. The present application solves the problems of ankle pump training, action monitoring, and thrombosis risk early warning for patients, and provides an interactive means and approach for guidance and encouragement between doctors and patients.

[0004] Based on the above-mentioned scheme, the limitations of the existing technology at least include the following problems. In the training risk identification process, the existing technology lacks dynamic modeling and process intervention capabilities for the user's training process. In actual training, users often produce small but critical action response deviations due to inattention or guidance understanding bias. The existing technology lacks modeling means for guidance alignment features, making it difficult to perceive such misalignment risks, which can easily cause high-risk states to be misjudged as normal achievements, thereby burying training accident hazards. Moreover, the existing technology does not establish a multi-source fusion analysis mechanism, which leads to inaccurate risk identification, and the intervention means lacks individual adaptability, thereby severely restricting the accuracy of training early warning. SUMMARY

[0005] In view of the deficiencies of the prior art, the present application provides an ankle pump training intelligent risk early warning interaction system and an intelligent ankle pump training device, which solve the problem that the prior art lacks multi-source fusion and is difficult to accurately identify risks in the training process and accurately warn.

[0006] To achieve the above object, the present application is implemented by the following technical solutions: an ankle pump training intelligent risk early warning interaction system, comprising: a data acquisition module for acquiring training data of a plurality of users, the training data including behavior data, physiological feedback data, and virtual target video stream data; a feature extraction module for analyzing behavior control ability characteristic values and physiological stress response characteristic values of each user based on the behavior data and the physiological feedback data of each user; a guidance analysis module for analyzing guide deviation characteristic values of each user based on virtual target video stream data of each user and in combination with a pre-trained guide response alignment identification model; a risk analysis module for analyzing training risk characteristic values of each user based on the behavior control ability characteristic values, the physiological stress response characteristic values, and the guide deviation characteristic values of each user; and an early warning interaction module for performing risk interaction early warning on each user based on the training risk characteristic values.

[0007] Further, the behavior data includes angle holding duration values, torque peak values, motion jitter values, force coordination values, muscle vibration fluctuation values, and foot pressure deviation values, and the specific steps of analyzing the behavior control ability characteristic values of each user are as follows: based on the behavior data of each user, analyzing behavior evaluation characteristic sets of each user, including stable control ability characteristic values and force balance performance characteristic values; and based on the behavior evaluation characteristic sets of each user, analyzing the behavior control ability characteristic values of each user.

[0008] Further, the specific steps of analyzing the behavior evaluation characteristic sets of each user are as follows: based on the angle holding duration values, the motion jitter values, and the muscle vibration fluctuation values of each user, analyzing the stable control ability characteristic values of each user; and based on the torque peak values, the force coordination values, and the foot pressure deviation values of each user, analyzing the force balance performance characteristic values of each user.

[0009] Further, the physiological feedback data includes foot temperature rise rate values, local blood flow change rate values, physiological variation values, foot dorsal blood oxygen fluctuation ratio values, and local physiological lag values, and the specific steps of analyzing the physiological stress response characteristic values of each user are as follows: based on the physiological feedback data of each user, analyzing physiological evaluation characteristic sets of each user, including local circulation adaptation characteristic values and physiological steady state regulation characteristic values; and based on the physiological evaluation characteristic sets of each user, analyzing the physiological stress response characteristic values of each user.

[0010] Further, the specific steps of analyzing the physiological evaluation feature set of each user are as follows: based on the foot temperature rise rate value, the local blood flow change rate value, and the local physiological lag value of each user, analyzing the local circulation adaptation feature value of each user; based on the physiological variation value and the dorsalis pedis blood oxygen fluctuation ratio of each user, analyzing the physiological steady state regulation feature value of each user.

[0011] Further, the virtual target video stream data includes a plurality of frames of virtual target image data, and each frame of virtual target image data includes a pixel point value and a two-dimensional coordinate of each pixel point in the virtual target image.

[0012] Further, the specific steps of analyzing the guidance deviation feature value of each user are as follows: inputting the virtual target video stream data of each user into the pre-trained guidance response alignment identification model, analyzing the target deviation feature set of each user, including the response stability feature value, the alignment fluctuation feature value, and the response direction concentration feature value; and based on the target deviation feature set of each user, analyzing the guidance deviation feature value of each user.

[0013] Further, the guidance response alignment identification model includes an input layer, a feature extraction layer, a time sequence association layer, and an output layer, and the specific steps of analyzing the target deviation feature set of each user are as follows: in the input layer of the guidance response alignment identification model, receiving each frame of virtual target image data of each user, and extracting a foot pixel point set and a virtual target pixel point set in each frame of virtual target image of each user; in the feature extraction layer of the guidance response alignment identification model, based on the foot region pixel point set and the virtual target pixel point set in each frame of virtual target image of each user, extracting an alignment feature vector in the corresponding frame of virtual target image; in the time sequence association layer of the guidance response alignment identification model, based on the alignment feature vector in each frame of virtual target image of each user, extracting a time sequence feature vector of each user; and in the output layer of the guidance response alignment identification model, performing output processing on the time sequence feature vector of each user to obtain the target deviation feature set of each user.

[0014] Further, the specific formula for calculating the training risk feature value of a certain user is as follows: ; wherein, , , , in turn are the training risk feature value, the behavior control ability feature value, the physiological stress response feature value, and the guidance deviation feature value of a certain user, , , , , in turn are the line control coefficient, the stress coefficient, the guidance deviation coefficient, the interaction coefficient, and the adjustment coefficient stored in the database.

[0015] The intelligent ankle pump trainer comprises a data acquisition unit for acquiring training data of a plurality of users, the training data comprising behavior data, physiological feedback data and virtual target video stream data; a behavior and physiology analysis unit for analyzing behavior control capability characteristic values and physiological stress response characteristic values of each user based on the behavior data and the physiological feedback data of each user; a guidance deviation analysis unit for analyzing guidance deviation characteristic values of each user based on the virtual target video stream data of each user and in combination with a pre-trained guidance response alignment identification model; a training risk analysis unit for analyzing training risk characteristic values of each user based on the behavior control capability characteristic values, the physiological stress response characteristic values and the guidance deviation characteristic values of each user; and a warning interaction feedback unit for performing risk interaction warning on each user based on the training risk characteristic values.

[0016] The present application has the following beneficial effects: (1) The ankle pump training intelligent risk warning interaction system fuses and acquires multi-source data of users through a data acquisition module, thereby establishing a unified multi-source input channel, extracting corresponding features in specific modules, finally extracting training risk characteristic values of each user by a risk analysis module, and performing warning feedback, so as to effectively realize joint perception and dynamic response of potential action instability, posture deviation and physiological abnormal reaction in the training process, thereby significantly improving the accuracy of risk identification, and helping to identify high-risk individuals in the early training and timely intervene, thereby avoiding safety hazards in the training process.

[0017] (2) The ankle pump training intelligent risk warning interaction system introduces a pre-trained guidance response alignment identification model, so as to perform pixel-level analysis on the foot and target area in each frame of virtual target image and extract time sequence alignment features, thereby effectively identifying slight action deviation of users in the training process caused by attention shift or response lag, thereby significantly enhancing the judgment ability of target response in the training process, and thereby helping the system to realize accurate deviation identification.

[0018] (3) The ankle pump training intelligent risk warning interaction system extracts corresponding characteristic values of each user based on the training data of each user, and generates training risk characteristic values based on the characteristic values, thereby ensuring that the system has good risk adaptability when facing different training stages or individual states, and the training risk characteristic values as the core driving index can trigger risk level warning prompts through the warning interaction module, thereby ensuring that the training system has active identification and response ability to high-risk states, and thereby comprehensively improving the intelligent level of training and the safety management ability of the system.

[0019] (4), the intelligent ankle pump trainer, through the cooperation of multiple units, to form a unit structure layout, thereby significantly improving the flexibility of the training equipment in the function deployment, the efficient collaborative operation of each functional unit is realized through the characteristic result conduction and the risk value driving mechanism, so that it has good scalability, and the structure design is helpful to realize the standardized integration between software and hardware components, facilitate rapid deployment and stable operation in various rehabilitation training environments, improve the practicability of the equipment.

[0020] Of course, implementing any product of the present application does not necessarily require achieving all the advantages described above at the same time. BRIEF DESCRIPTION OF DRAWINGS

[0021] Figure 1 The ankle pump training intelligent risk early warning interaction system block diagram of the present application.

[0022] Figure 2 The user sequence data schematic diagram of the behavior evaluation feature set in the ankle pump training intelligent risk early warning interaction system of the present application.

[0023] Figure 3 The specific step flow chart of analyzing the target deviation feature set of each user in the ankle pump training intelligent risk early warning interaction system of the present application.

[0024] Figure 4 The intelligent ankle pump training device block diagram of the present application. DETAILED DESCRIPTION

[0025] Please refer to Figure 1 The embodiment of the present application provides a technical scheme: an ankle pump training intelligent risk early warning interaction system, comprising: a data acquisition module, used for acquiring training data of a plurality of users within a set period (such as 10s), the training data including behavior data, physiological feedback data, and virtual target video stream data; a feature extraction module, used for analyzing behavior control ability characteristic values and physiological stress response characteristic values of each user based on the behavior data and the physiological feedback data of each user; a guidance analysis module, used for analyzing guide deviation characteristic values of each user based on the virtual target video stream data of each user and combining a pre-trained guide response alignment identification model; a risk analysis module, used for analyzing training risk characteristic values of each user based on the behavior control ability characteristic values, the physiological stress response characteristic values, and the guide deviation characteristic values of each user; and a warning interaction module, used for performing risk interaction warning on each user based on the training risk characteristic values, which specifically comprises: The training risk characteristic value of each user is respectively judged and processed with the preset risk interval. If the training risk characteristic value of each user is lower than the lower limit of the preset risk interval, the setting period of the user is marked as low risk, and no early warning is triggered. If the training risk characteristic value of each user is within the preset risk interval, the setting period of the user is marked as medium risk, and a voice broadcast or a text reminder on the training interface is performed. If the training risk characteristic value of each user is higher than the upper limit of the preset risk interval, the setting period of the user is marked as high risk, and a high-intensity early warning mechanism is immediately triggered, including pausing the current training process, issuing a highlighted sound and light alarm signal, and popping up a risk prompt window on the training interface to prompt the user to pause training and seek professional guidance.

[0026] The specific formula for calculating the training risk characteristic value of a user is as follows: ; wherein, is the training risk characteristic value of a user, is the behavior control ability characteristic value of a user, is the line control coefficient stored in the database, is the physiological stress response characteristic value of a user, is the stress coefficient stored in the database, is the guidance deviation characteristic value of a user, is the deviation coefficient stored in the database, is the interaction coefficient stored in the database, is the adjustment coefficient stored in the database, and in the present embodiment example, the value is 3.000.

[0027] It needs to be explained that the line control coefficient stored in the database is The acquisition steps are as follows: reading the behavior control ability characteristic value, the physiological stress response characteristic value, and the guidance deviation characteristic value of each user, and performing mean value processing to obtain the behavior control ability characteristic mean value (the value takes the reciprocal), the physiological stress response characteristic mean value, and the guidance deviation characteristic mean value of each user, and performing summation processing to obtain a risk sum value, performing ratio processing on the behavior control ability characteristic mean value of each user and the risk sum value, and taking the result as the line control coefficient ; The stress coefficient stored in the database is The acquisition steps are as follows: reading the behavior control ability characteristic mean value (the value takes the reciprocal), the physiological stress response characteristic mean value, and the guidance deviation characteristic mean value of each user, and performing summation processing to obtain a risk sum value, and performing ratio processing on the physiological stress response characteristic mean value of each user and the risk sum value, and taking the result as the stress coefficient ; The deviation coefficient stored in the database is The acquisition step is as follows: read the mean value of each user's behavior control ability characteristic (the value is taken as the reciprocal), the mean value of the physiological stress response characteristic, and the mean value of the guidance deviation characteristic, and perform summation processing to obtain a risk sum value, perform ratio processing on the mean value of each user's guidance deviation characteristic and the risk sum value, and take the result as the guidance deviation coefficient ; The interaction coefficient stored in the database The acquisition step is as follows: read the mean value of each user's behavior control ability characteristic (the value is taken as the reciprocal), the mean value of the physiological stress response characteristic, and the mean value of the guidance deviation characteristic, and perform summation processing to obtain a risk sum value, perform ratio processing on the mean value of each user's guidance deviation characteristic and the risk sum value, and take the result as the guidance deviation coefficient .

[0028] Specifically, the behavior data includes angle holding time value, torque peak value, motion jitter value, force coordination value, muscle vibration fluctuation value, and foot pressure skewness value. The specific steps for analyzing the behavior control ability characteristic value of each user are as follows: based on the behavior data of each user, analyze the behavior evaluation characteristic set of each user, including stable control ability characteristic value and force balance performance characteristic value; based on the behavior evaluation characteristic set of each user, analyze the behavior control ability characteristic value of each user.

[0029] The angle holding time value is the time length that each user can continuously maintain the ankle joint angle within a set range during the training period, reflecting the motion control ability and continuous stability of each user. The angle value of each time point of the ankle joint is collected by a 9-axis inertial measurement unit (IMU), and whether the user can stably maintain within the set angle range is monitored during the training process. When the angle change is less than a preset threshold (such as ±2°), the timer starts to work until the angle change exceeds the threshold or exceeds the set time limit.

[0030] The torque peak value is the maximum torque value that each user can apply to the ankle joint during the training process in a set period. The maximum torque value of the left and right ankle joints at each time point is collected by a 9-axis inertial measurement unit (IMU), and the maximum torque values of the left and right ankle joints are taken respectively, and the mean value is taken as the torque peak value.

[0031] The motion jitter value is the fluctuation degree of the ankle joint angle during the training process of each user in a set period, reflecting the motion stability and control accuracy. The angle value of each time point of the ankle joint is collected in real time by a 9-axis inertial measurement unit (IMU), and the angle variance value is taken as the motion jitter value.

[0032] The force coordination value is the coordination of the left and right ankle forces of each user, reflecting whether there is asymmetric force of the bilateral ankles, evaluating the balance and stability of the user's action, which can be obtained by the 9-axis inertial measurement unit (IMU) to collect the torque value of the left and right ankles at each time point, and respectively difference processing (i.e. the absolute value of the difference between the torque values of the left and right ankles), to obtain the torque difference value at each time point, and to perform mean value processing to obtain the torque coordination value, and through the dual-foot pressure sensor array to obtain the pressure value of each position of the left and right feet at each time point, and to perform mean value processing to obtain the pressure mean value of the left and right feet, and to difference processing, and to standardize the results with the torque coordination value, and to weight based on the results to obtain the force coordination value.

[0033] The muscle vibration fluctuation value is the change amplitude of muscle micro-vibration, reflecting the microscopic activity state of muscle, helping to judge muscle fatigue or movement abnormalities, which can be obtained by the micro-vibration sensor to obtain the muscle vibration frequency value at each time point, and to perform variance processing, and to take the results as the muscle vibration fluctuation value.

[0034] The foot pressure skewness value is a measure of the skewness of the foot pressure distribution, which can be obtained by the foot pressure sensor array to obtain the pressure value of each position of the left and right feet at each time point, and based on the sample skewness formula to analyze the skewness value of the left and right feet at each time point, and to extract the skewness mean value of the left and right feet, and to weight processing, and to take the results as the foot pressure skewness value.

[0035] The specific formula for calculating the behavior control ability characteristic value of a user is as follows: ; wherein, is the behavior control ability characteristic value of a user, is the stable control ability characteristic value of a user, is the stable control coefficient stored in the database, is the force balance performance characteristic value of a user, is the balance coefficient stored in the database, .

[0036] It needs to be explained that the stable control coefficient and the balance coefficient stored in the database are obtained as follows: read the stable control ability characteristic value and the force balance performance characteristic value of each user, and perform mean value processing to obtain the stable control ability characteristic mean value and the force balance performance characteristic mean value of each user, and perform summation processing to obtain the behavior control sum value, and perform ratio processing of the stable control ability characteristic mean value and the force balance performance characteristic mean value of each user with the behavior control sum value, and take the results as the stable control coefficient and the balance coefficient .

[0037] The specific implementation example of calculating the behavior control ability characteristic value of a certain user is as follows, and the following data is provided, including the stable control ability characteristic values and the force balance performance characteristic values of 5 users (randomly selected), as shown in Table 1 and Figure 2 Table 1. Example of behavior evaluation characteristic set user sequence database The stable control coefficient stored in the database is about 0.584. The balance coefficient stored in the database is about 0.416. The balance coefficient stored in the database is about 0.416. Substitute the data in Table 1 and the above coefficients into the specific formula for calculating the behavior control ability characteristic value of a certain user to obtain: The behavior control ability characteristic value of the first user is 0.584 x 0.734 + 0.416 x 0.667 ≈ 0.706. The behavior control ability characteristic value of the second user is 0.584 x 0.683 + 0.416 x 0.643 ≈ 0.666. The behavior control ability characteristic value of the third user is 0.584 x 0.763 + 0.416 x 0.713 ≈ 0.742. The behavior control ability characteristic value of the fourth user is 0.584 x 0.741 + 0.416 x 0.694 ≈ 0.721. The behavior control ability characteristic value of the fifth user is 0.584 x 0.654 + 0.416 x 0.637 ≈ 0.647.

[0038] The specific steps for analyzing the behavior evaluation characteristic set of each user are as follows: based on the angle holding time value, motion jitter value, and muscle vibration fluctuation value of each user, the stable control ability characteristic value of each user is analyzed, which is specifically: the angle holding time value, motion jitter value, and muscle vibration fluctuation value of each user are standardized, and the weighted processing result is taken as the stable control ability characteristic value; based on the torque peak value, force coordination value, and foot pressure deviation value of each user, the force balance performance characteristic value of each user is analyzed, which is specifically: the torque peak value, force coordination value, and foot pressure deviation value of each user are standardized, and the weighted processing result is taken as the force balance performance characteristic value.

[0039] ​​In this embodiment, the hierarchical mapping path of the stability control ability and the balanced performance is constructed by constructing the behavior data, so that the original complex multi-dimensional action data is structured into evaluation indicators with physical meaning, facilitating the system to understand the characteristics and model the risk. Secondly, the structure design classifies multiple action parameters into stability and coordination, and effectively weakens the parameter scale deviation problem caused by individual differences through standardization and weighted fusion, thereby improving the adaptability of the system to different user behavior patterns. Finally, the stability control coefficient and the balance coefficient are generated, which can realize the comparative evaluation of individual characteristics and group baseline, and provide a comparable basis for subsequent modeling of training risk, thereby helping to build a risk warning for actual training scenarios.

[0040] Specifically, the physiological feedback data includes foot temperature rise rate value, local blood flow change rate value, physiological variation value, foot dorsal blood oxygen fluctuation ratio value, and local physiological lag value. The specific steps of analyzing the physiological stress response characteristic value of each user are as follows: based on the physiological feedback data of each user, the physiological evaluation characteristic set of each user is analyzed, including local circulation adaptation characteristic value and physiological steady state regulation characteristic value; based on the physiological evaluation characteristic set of each user, the physiological stress response characteristic value of each user is analyzed, which is specifically: the local circulation adaptation characteristic value and the physiological steady state regulation characteristic value of each user are weighted and processed, and a ratio processing is performed, i.e. 1 / (1+weighted processing result), to obtain the physiological stress response characteristic value of each user.

[0041] The foot temperature rise rate value is the change speed of the foot surface temperature during the training process, which reflects whether the microcirculation of each user's foot is effectively activated and the change trend of local metabolic capacity. It can be obtained by a thermistor array distributed in the foot area, and the temperature value of each position at each time point in a set period is obtained, and the average foot temperature at the start and end of the period is extracted, and the difference is analyzed, and the period length is processed by ratio, and the result is taken as the foot temperature rise rate value.

[0042] The local blood flow change rate value is the dynamic change speed of the foot microcirculation blood flow during the training process, which reflects the vascular dilation effect and venous return efficiency. It can be collected by a built-in flexible microfluidic sensing array to collect the instantaneous blood flow signal of multiple areas of the foot, and local integral processing is performed at each time point, and the average blood flow variation rate is calculated in a set period.

[0043] The physiological variation value is the overall fluctuation amplitude of multiple physiological parameters during the training process, which reflects the comprehensive regulation stability of the neural-vascular system of each user. It can be obtained by synchronously collecting the dorsalis pedis blood oxygen value, foot temperature value, and blood flow value at each time point in a set period, and extracting the standard deviation value of the response, and summing the square of the standard deviation and taking the square root, and taking the result as the physiological variation value.

[0044] The dorsum pedis blood oxygen fluctuation ratio is a characteristic value for representing the fluctuation ability of the foot blood oxygen level during the training process, and is used to reflect the oxygenation response strength of the peripheral circulation of each user under the training intervention. The dorsum pedis blood oxygen value at each time point can be obtained by installing an integrated photoelectric blood oxygen sensor (using the red light / infrared ratio method) in the dorsum pedis area, and the average dorsum pedis blood oxygen value and the peak dorsum pedis blood oxygen value are extracted respectively, and the ratio is processed, and the result is taken as the dorsum pedis blood oxygen fluctuation ratio.

[0045] The local physiological lag value is a response time difference reflecting the response sequence difference of foot blood flow improvement and temperature rise, which can reveal the degree of synchronization and coordination of the microvascular and nerve regulation system. The time when the dorsum pedis blood oxygen saturation reaches a set threshold (such as an increase of 2%) after the start of training and the time required for the foot temperature to rise by 1.5°C are recorded respectively, and the response time difference (absolute value) is calculated, and the result is taken as the local physiological lag value.

[0046] The specific steps for analyzing the physiological evaluation characteristic set of each user are as follows: based on the foot temperature rise rate value, the local blood flow change rate value, and the local physiological lag value of each user, the local circulation adaptation characteristic value of each user is analyzed, which is specifically: the foot temperature rise rate value, the local blood flow change rate value, and the local physiological lag value of each user are standardized, and the result is taken as the local circulation adaptation characteristic value based on the standardized processing result; based on the physiological variation value and the dorsum pedis blood oxygen fluctuation ratio of each user, the physiological steady state regulation characteristic value of each user is analyzed, which is specifically: the physiological variation value and the dorsum pedis blood oxygen fluctuation ratio of each user are standardized, and the result is taken as the physiological steady state regulation characteristic value based on the standardized processing result.

[0047] In this embodiment, by finely disassembling the physiological feedback data and dividing it into intermediate indicators with clear physiological significance, the structured understanding and modeling ability of the individual physiological stress state is significantly enhanced, and the microcirculation activation level and neurovascular coordinated regulation of the user under the training stimulus are comprehensively reflected. Secondly, through standardized and weighted fusion processing, the natural physiological differences of different users are effectively weakened, thereby improving the adaptability and robustness of the system across users, and reflecting the fine perception of the micro physiological dynamic process. Finally, the evaluation channels of the local circulation adaptation characteristic value and the physiological steady state regulation characteristic value are constructed, so that the physiological feedback data is upgraded to a structured cognitive indicator with predictive value. Especially in the dynamic training scene, this modeling method can realize early identification and trend prediction of abnormal stress response, which helps to discover potential risks in time and give early warnings, and ultimately ensures the safety of the whole process of ankle pump training.

[0048] Specifically, the virtual target video stream data includes a plurality of frames of virtual target image data (containing a mixed picture of each user's foot and the projected virtual target), and each frame of virtual target image data includes a pixel point value and a two-dimensional coordinate of each pixel point in the virtual target image.

[0049] The specific steps of analyzing the guiding deviation feature value of each user are as follows: inputting the virtual target video stream data of each user into the pre-trained guiding response alignment identification model, analyzing the target deviation feature set of each user, including the response stability feature value, the alignment fluctuation feature value and the response direction concentration feature value; and based on the target deviation feature set of each user, analyzing the guiding deviation feature value of each user, which is specifically: performing weighted processing on the response stability feature value, the alignment fluctuation feature value and the response direction concentration feature value of each user, and taking the result as the guiding deviation feature value.

[0050] As shown in Figure 3 The specific steps of analyzing the target deviation feature set of each user are as follows: in the input layer of the guiding response alignment identification model, receiving each frame of virtual target image data of each user, and extracting a foot pixel point set and a virtual target pixel point set in each frame of virtual target image of each user, which is specifically: using a lightweight semantic segmentation neural network (such as U-Net, BiSeNet, etc.) to perform segmentation identification on each frame of virtual target image, to obtain left and right foot regions and left and right virtual target regions, and marking the pixel points in the corresponding regions, such as marking each pixel point in the left foot region as a corresponding left foot pixel point, marking a plurality of left foot pixel points and a plurality of right foot pixel points as the foot pixel point set, and marking a plurality of left target pixel points and a plurality of right target pixel points as the virtual target pixel point set; In the feature extraction layer of the guiding response alignment identification model, based on the foot region pixel point set and the virtual target pixel point set in each frame of virtual target image of each user, an alignment feature vector in the corresponding frame of virtual target image is extracted; in the time sequence correlation layer of the guiding response alignment identification model, based on the alignment feature vector in each frame of virtual target image of each user, a time sequence feature vector of each user is extracted, which is specifically: the LSTM network processes the time dependence of the data, captures the long-term dependence between each time step by learning, and captures the time change trend in the data, in the LSTM layer, the network updates the hidden state for each time step through the mechanism of recurrent neural network (RNN), thereby remembering the information in the time sequence, and extracting the features of the time sequence according to the information, this process can process the change pattern of each feature over time, and generate time sequence features that can reflect these time changes, such as: extracting distance deviation mean features and distance deviation peak features in the alignment feature vector, and performing ratio processing, i.e. distance deviation mean features / distance deviation peak features, to extract a response stability ratio feature representing the response degree of each user to the virtual target during the training process in the set period; extracting an IoU overlap feature in the alignment feature vector, and performing standard deviation processing to extract an alignment fluctuation feature representing the spatial consistency of each user with the guiding target during the training process in the set period; extracting the barycentric two-dimensional coordinates of the left and right foot regions and the left and right virtual target regions in the alignment feature vector, and extracting a left deviation vector (the difference between the barycentric two-dimensional coordinates of the left foot region and the left virtual target region), performing ellipse fitting on the left deviation vector in each frame of virtual target image, performing fitting parameter estimation (such as the direct ellipse fitting algorithm proposed by Fitzgibbon et al.) using the least squares method, solving the general form of the standard equation of the ellipse, and calculating the major axis radius and the minor axis radius of the ellipse through matrix eigenvalue decomposition, and performing ratio processing, i.e. √(1-(minor axis radius squared / major axis radius squared)), to extract a left eccentricity feature, and similarly obtaining a right eccentricity feature, and performing weighted processing to extract a response direction concentration feature representing the direction concentration of each user during the training process in the set period, and the more concentrated the direction is, the more accurate the control is; and performing splicing processing on the response stability ratio feature, the alignment fluctuation feature, and the response direction concentration feature to form a time sequence feature vector in each frame of virtual target image; In the output layer of the guiding response alignment recognition model, the time sequence feature vector of each user is outputted to obtain a target deviation feature set of each user, which specifically includes: outputting the response stability ratio feature, the alignment fluctuation feature, and the response direction concentration feature in the time sequence feature vector into specific numerical values (such as Sigmoid function, etc.), including the response stability feature value, the alignment fluctuation feature value, and the response direction concentration feature value.

[0051] The specific steps of extracting the feature vector in each frame of virtual target image of each user are as follows: reading the two-dimensional coordinates of each left foot pixel point and the two-dimensional coordinates of each left target pixel point in each frame of virtual target image of each user, performing mean processing respectively to extract the barycentric two-dimensional coordinates of the left foot region and the left virtual target region, and extracting the left foot distance deviation value based on the Euclidean distance formula, and similarly obtaining the right foot distance deviation value, and performing weighted processing to extract the distance deviation feature. The two-dimensional coordinates of each left foot pixel point and the two-dimensional coordinates of each left target pixel point in each frame of virtual target image of each user are read, and a minimum bounding box fitting process is performed respectively to obtain a left foot region minimum bounding box and a left virtual target region minimum bounding box, and both are represented by the upper left corner coordinates and the lower right corner coordinates. The intersection area and the union area are extracted, that is, the width of the intersection region, max[0, min (the X-axis coordinate value of the lower right corner of the left foot region minimum bounding box, the X-axis coordinate value of the lower right corner of the left virtual target region minimum bounding box) - max (the X-axis coordinate value of the lower left corner of the left foot region minimum bounding box, the X-axis coordinate value of the lower left corner of the left virtual target region minimum bounding box)], the height of the intersection region, max[0, min (the Y-axis coordinate value of the lower right corner of the left foot region minimum bounding box, the Y-axis coordinate value of the lower right corner of the left virtual target region minimum bounding box) - max (the Y-axis coordinate value of the lower left corner of the left foot region minimum bounding box, the Y-axis coordinate value of the lower left corner of the left virtual target region minimum bounding box)], and the intersection area is obtained by multiplying them. The area of the left foot region minimum bounding box is extracted, that is, (the X-axis coordinate value of the lower right corner of the left foot region minimum bounding box - the X-axis coordinate value of the lower left corner of the left foot region minimum bounding box) x (the Y-axis coordinate value of the lower right corner of the left foot region minimum bounding box - the Y-axis coordinate value of the lower left corner of the left foot region minimum bounding box), and the area of the left virtual target region minimum bounding box is extracted in the same way. The sum of the areas of the left foot region minimum bounding box and the left virtual target region minimum bounding box is calculated, and the difference between the result and the intersection area is processed to extract the union area. The intersection area and the union area are processed by ratio, that is, the intersection area / the union area, to obtain the left foot overlap feature. The right foot overlap feature is obtained in the same way, and the weighted processing is performed to extract the IoU overlap feature. The distance deviation feature, the IoU overlap feature, and the barycentric two-dimensional coordinates of the left and right foot regions and the left and right virtual target regions are spliced to form an alignment feature vector in each frame of virtual target image.

[0052] The pre-training step of the guided response alignment recognition model is as follows: In the pre-training phase of the guided response alignment recognition model, a virtual target interaction training set containing multiple user postures and response behavior features is first constructed. The training set includes a large number of virtual target image frame data generated by multiple users of different genders, ages, and response abilities in standard training guide tasks and their supporting annotations. The annotation content includes: the pixel mask of the left and right foot regions in each frame of image, the pixel mask of the left and right virtual target regions, the expected direction vector value, the foot response trajectory, the target deviation level label, etc.

[0053] For the collected image data, the following preprocessing steps are performed: normalization (brightness, size, coordinate system), artifact removal and size scaling; image enhancement strategies (such as rotation, occlusion simulation, background change, etc.) are used to improve the robustness of the model to complex use environments.

[0054] After completing data preparation, the model input layer, feature extraction layer, and time series association layer are sequentially pre-trained: input layer pre-training, input is each frame of virtual target image, label is the pixel mask of foot area and virtual target area, using lightweight semantic segmentation network (such as U-Net, BiSeNet), output the pixel set of left and right foot area and left and right target area in each frame image, the training loss function uses class weighted cross entropy, focusing on enhancing the classification accuracy of boundary pixels, the trained model can automatically extract the left and right foot area and target area from any user virtual target interaction image; feature extraction layer pre-training: input is the center of gravity coordinates and the outer rectangle coordinates of the left and right foot and target area in each frame image, the alignment feature vector of each frame is extracted by rule logic calculation, the supervision signal of this layer is the annotated target deviation level value (multi-level), the loss function uses mean square error loss and center offset constraint term to ensure that the extracted features can effectively distinguish the response ability; Time series association layer pre-training: input is the sequence of alignment feature vectors of consecutive frames, use double-layer LSTM network for time series modeling, the training target is to identify the overall response deviation feature trend in each training period, including response stability ratio (deviation mean / peak), IoU fluctuation feature (IoU standard deviation), response direction concentration (deviation ellipse fitting eccentricity), the supervision signal is the overall behavior performance level annotated by clinical experts corresponding to each training period, Adam optimizer (initial learning rate 1e-4, weight decay 1e-5) is used in training, Dropout mechanism (p=0.3) is added to prevent overfitting, and early stopping strategy is set to save the optimal model.

[0055] After the final training is completed, the input layer, feature extraction layer, and time series association layer are sequentially assembled into a complete guided response alignment recognition model and deployed in the training system to identify the deviation degree and direction control features of user virtual target response behavior in real time, providing core support indicators for ankle pump training risk warning system.

[0056] In the embodiment, the dynamic response deviation between the user's foot action and the guide target is structured modeled by constructing a guide response alignment identification model, which significantly enhances the system's ability to identify behavior mismatch and direction control abnormalities. Secondly, through each layer in the guide response alignment identification model, the model extracts the pixel-level boundary information of the foot and the target area from the original virtual target image. Through the Euclidean distance, coincidence degree analysis and deviation ellipse fitting, the core feature vector reflecting the response accuracy and direction concentration is quantitatively generated. By introducing LSTM in the time sequence layer, the stability and fluctuation trend of the user's response behavior can be deeply mined. Finally, during the model pre-training stage, multi-age and multi-state user samples are introduced, and image enhancement and label supervision mechanisms are introduced, which effectively improves the clinical adaptability of the model and ensures its stable operation under actual complex training conditions, thereby improving the system's early identification ability of high-risk response behavior.

[0057] Referring to Figure 4 The embodiment of the present application provides a technical solution: an intelligent ankle pump trainer, comprising: a data acquisition unit for acquiring training data of a plurality of users, the training data including behavior data, physiological feedback data, and virtual target video stream data; a behavior and physiology analysis unit for analyzing behavior control ability characteristic values and physiological stress response characteristic values of each user based on the behavior data and the physiological feedback data of each user; a guide deviation analysis unit for analyzing guide deviation characteristic values of each user based on the virtual target video stream data of each user and in combination with a pre-trained guide response alignment identification model; a training risk analysis unit for analyzing training risk characteristic values of each user based on the behavior control ability characteristic values, the physiological stress response characteristic values, and the guide deviation characteristic values of each user; and a warning interaction feedback unit for performing risk interaction warning on each user based on the training risk characteristic values.

[0058] Although preferred embodiments of the application have been described, those skilled in the art, once aware of the basic inventive concept, can make additional changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the present application.

[0059] Obviously, those skilled in the art can make various modifications and variations to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalent technologies, the present application also intends to include these modifications and variations.

Claims

1. An intelligent risk warning and interactive system for ankle pump training, characterized in that, include: The data acquisition module is used to acquire training data from several users, including behavioral data, physiological feedback data, and virtual target video stream data. The virtual target video stream data includes several frames of virtual target image data, and each frame of virtual target image data includes the pixel value and two-dimensional coordinates of each pixel in the virtual target image; The feature extraction module receives behavioral data and physiological feedback data acquired by the data acquisition module. The feature extraction module processes and analyzes the acquired behavioral data and physiological feedback data of each user to obtain behavioral control ability feature values ​​and physiological stress response feature values ​​of each user. The guidance analysis module receives virtual target video stream data acquired by the data acquisition module. The guidance analysis module analyzes the acquired virtual target video stream data of each user in combination with the trained guidance response alignment recognition model to obtain the guidance deviation feature value of each user. The risk analysis module receives control ability feature values, physiological stress response feature values, and guidance deviation feature values ​​obtained by the feature extraction module and the guidance analysis module, respectively. The risk analysis module processes and analyzes the behavioral control ability feature values, physiological stress response feature values, and guidance deviation feature values ​​of each user to analyze the training risk feature values ​​of each user. The early warning and interaction module provides risk interaction and early warning to each user based on trained risk feature values.

2. The ankle pump training intelligent risk warning and interactive system according to claim 1, characterized in that, The behavioral data includes angle holding time, peak torque, motion jitter, force coordination, muscle vibration fluctuation, and foot pressure deviation. The specific steps for analyzing each user's behavioral control ability characteristics are as follows: Based on each user's behavioral data, analyze each user's behavioral evaluation feature set, including stable control ability feature value and balanced force performance feature value; Based on each user's behavioral assessment feature set, we analyze the behavioral control ability feature values ​​of each user.

3. The ankle pump training intelligent risk warning and interactive system according to claim 2, characterized in that, The specific steps for analyzing each user's behavioral evaluation feature set are as follows: Based on each user's angle holding time, motion jitter, and muscle vibration fluctuation, the stability control ability characteristics of each user are analyzed. Based on each user's peak torque, force coordination value, and foot pressure deviation value, the characteristic values ​​of each user's force balance performance are analyzed.

4. The ankle pump training intelligent risk warning and interactive system according to claim 1, characterized in that, The physiological feedback data includes foot temperature rise rate, local blood flow change rate, physiological variability, dorsolateral foot blood oxygen fluctuation ratio, and local physiological lag value. The specific steps for analyzing the physiological stress response characteristics of each user are as follows: Based on each user's physiological feedback data, analyze each user's physiological assessment feature set, including local circulatory adaptation feature values ​​and physiological homeostasis regulation feature values; Based on each user's physiological assessment feature set, the physiological stress response feature values ​​of each user are analyzed.

5. The ankle pump training intelligent risk warning and interactive system according to claim 4, characterized in that, The specific steps for analyzing each user's physiological assessment feature set are as follows: Based on each user's foot temperature rise rate, local blood flow change rate, and local physiological hysteresis value, the local circulatory adaptation characteristics of each user were analyzed. Based on each user's physiological variability and foot dorsolateral blood oxygen fluctuation ratio, the physiological homeostasis regulation characteristics of each user were analyzed.

6. The ankle pump training intelligent risk warning and interactive system according to claim 1, characterized in that, The specific steps for analyzing the guidance deviation characteristic values ​​of each user are as follows. The virtual target video stream data of each user is input into the pre-trained guided response alignment recognition model to analyze the target deviation feature set of each user, including response stability ratio feature value, alignment fluctuation feature value, and response direction concentration feature value. Based on each user's target deviation feature set, the guidance deviation feature value of each user is analyzed.

7. The ankle pump training intelligent risk warning and interactive system according to claim 6, characterized in that, The guided response alignment recognition model includes an input layer, a feature extraction layer, a temporal correlation layer, and an output layer. The specific steps for analyzing the target deviation feature set of each user are as follows: In the input layer of the guided response alignment recognition model, the virtual target image data of each user for each frame is received, and the foot pixel set and virtual target pixel set in each user's virtual target image for each frame are extracted. In the feature extraction layer of the guided response alignment recognition model, the alignment feature vector in the corresponding frame of the virtual target image is extracted based on the set of pixels in the foot region and the set of pixels in the virtual target image of each user in each frame of the virtual target image. In the temporal correlation layer of the guided response alignment recognition model, the temporal feature vector of each user is extracted based on the alignment feature vector in each frame of the virtual target image of each user. In the output layer of the guided response alignment recognition model, the temporal feature vector of each user is processed to obtain the target deviation feature set of each user.

8. The ankle pump training intelligent risk warning and interactive system according to claim 7, characterized in that, The specific formula for calculating the training risk feature value of a user is as follows: ； in, , , , The following are, in order, a user's training risk characteristic value, behavioral control ability characteristic value, physiological stress response characteristic value, and guidance deviation characteristic value. , , , , The coefficients stored in the database are, in order: control coefficient, stress coefficient, bias coefficient, interaction coefficient, and adjustment coefficient.

9. An intelligent ankle pump trainer, employing the intelligent risk warning and interactive system for ankle pump training as described in any one of claims 1-8, characterized in that, include: The data acquisition unit is used to acquire training data from several users, including behavioral data, physiological feedback data, and virtual target video stream data. The behavioral and physiological analysis unit is used to analyze each user's behavioral control ability characteristics and physiological stress response characteristics based on each user's behavioral data and physiological feedback data. The guidance deviation analysis unit is used to analyze the guidance deviation feature value of each user based on the virtual target video stream data of each user and combined with the pre-trained guidance response alignment recognition model. The training risk analysis unit is used to analyze the training risk characteristics of each user based on their behavioral control ability characteristics, physiological stress response characteristics, and guidance deviation characteristics. The early warning and interactive feedback unit provides risk interactive early warnings to each user based on trained risk feature values.

Images