Rehabilitation training optimization method, system and equipment based on artificial intelligence and medium

Abstract

The invention relates to a rehabilitation training optimization method and system based on artificial intelligence, equipment and a medium. The method comprises the following steps: acquiring basic physiological data of a user; the basic physiological data at least comprises a heart injury reason, an age parameter, a heart rate parameter, a heart pressure parameter and a body mass index; generating an initialized exercise plan of the user based on the basic physiological data; when the initialized exercise plan is executed, acquiring exercise image data, real-time heart rate parameters, fatigue image data and fatigue voice data of the user; quantifying the effect of the initialized exercise plan on the user based on the exercise image data, the real-time heart rate parameter, the fatigue feeling image data and the fatigue feeling voice data to obtain plan feedback of the initialized exercise plan; and updating the initialized exercise plan by using plan feedback and / or historical training data of the user to obtain an updated initialized exercise plan. By adopting the method, the heart rehabilitation training efficiency can be improved.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of cardiac rehabilitation training, and particularly relates to a rehabilitation training optimization method, system, device and medium based on artificial intelligence. BACKGROUND

[0002] With the development of rehabilitation training technology, an intelligent rehabilitation monitoring system based on wearable devices appears, which can continuously collect basic physiological data such as heart rate and blood pressure of patients and realize remote transmission, thereby forming the current mainstream "periodic assessment-static prescription-self-execution" cardiac rehabilitation mode.

[0003] In the existing intelligent rehabilitation monitoring system based on wearable devices, a cardiac rehabilitation physician first formulates an initialization rehabilitation scheme including exercise type, intensity, frequency and duration based on the electronic health record and the initial assessment result of the patient; the patient performs training according to the established scheme in the home environment, records the heart rate change through a smart bracelet and manually fills in the training log; the physician obtains the training data and subjective feeling of the patient through video follow-up or outpatient review, judges the rehabilitation progress combined with clinical experience, and adjusts the exercise prescription in stages such as appropriately increasing the exercise intensity or prolonging the single training time in units of weeks or months.

[0004] However, the current rehabilitation method has obvious limitations: first, the static exercise prescription is difficult to adapt to the daily physiological state and rehabilitation needs of the patient, and lacks a targeted dynamic adjustment mechanism; second, artificial assessment depends on the experience of the physician and has subjective bias, and cannot realize precise quantitative analysis; third, and most importantly, the traditional method lacks real-time monitoring and immediate intervention ability in the training execution process, and it is difficult to correct when the patient makes posture errors or is excessively tired, which affects the rehabilitation effect and has safety hazards, resulting in low cardiac rehabilitation training efficiency. SUMMARY

[0005] Therefore, it is necessary to provide a rehabilitation training optimization method, system, device and medium based on artificial intelligence which can improve the cardiac rehabilitation training efficiency in view of the above technical problems.

[0006] In a first aspect, the application provides a rehabilitation training optimization method based on artificial intelligence, comprising:

[0007] obtaining basic physiological data of a user; the basic physiological data at least includes a heart damage reason, an age parameter, a heart rate parameter, a heart pressure parameter and a body mass index;

[0008] generating an initialization exercise plan of the user based on the basic physiological data; the initialization exercise plan is used to indicate an exercise type, an exercise frequency, an exercise time and an exercise heart rate interval of the user exercise;

[0009] In the process of executing the initial exercise plan, the exercise image data, real-time heart rate parameters, fatigue image data and fatigue voice data of the user are acquired;

[0010] Based on the exercise image data, real-time heart rate parameters, fatigue image data and fatigue voice data, the effect of the initial exercise plan on the user is quantified to obtain plan feedback of the initial exercise plan;

[0011] Using the plan feedback and / or historical training data of the user, the initial exercise plan is updated to obtain an updated initial exercise plan.

[0012] Further, based on the basic physiological data, the initial exercise plan of the user is generated, including:

[0013] Based on the preset exercise intensity parameters, combined with the age parameters and heart rate parameters, the following formula is used to calculate the exercise heart rate interval data:

[0014]

[0015] wherein, is the exercise heart rate interval, is the heart rate parameter, is the first exercise intensity parameter, is the age parameter, is the second exercise intensity parameter;

[0016] The heart damage reason and body mass index are input into the preset exercise type classification model to obtain exercise type data; the exercise type classification model is constructed based on a random forest algorithm;

[0017] The heart pressure parameter and the body mass index are input into the preset multiple linear regression model to obtain exercise frequency data and exercise time data;

[0018] Based on the exercise heart rate interval data, exercise type data, exercise frequency data and exercise time data, the initial exercise plan is obtained.

[0019] Further, based on the exercise image data, real-time heart rate parameters, fatigue image data and fatigue voice data, the effect of the initial exercise plan on the user is quantified to obtain plan feedback of the initial exercise plan, including:

[0020] Based on the preset standard exercise key point sequence and the exercise image data, the difference between the actual exercise of the user and the correct exercise of the user is quantified to obtain posture difference data; the posture difference data includes posture angle change data and posture key point change data; the standard exercise key point sequence is used to represent the standard posture of the user during exercise;

[0021] based on the gesture key point change data, quantifying the difference between the coordinates of the key points in the user's movement and the coordinates of the key points in the standard gesture, obtaining a gesture correctness score;

[0022] based on the gesture angle change data, the gesture key point change data and the standard movement key point sequence, generating auditory feedback information; the auditory feedback information is used to instruct the user to adjust the training gesture through auditory information;

[0023] based on the fatigue feeling image data and the fatigue feeling voice data, predicting the fatigue degree of the user, obtaining a comprehensive fatigue score;

[0024] based on the real-time heart rate parameter and the initial movement plan, quantifying the normal degree of heart rate in the user's movement, obtaining a heart rate compliance degree;

[0025] According to the preset weighting weight, the gesture correctness score and the heart rate compliance degree are weighted and fused to obtain a movement success rate;

[0026] based on the auditory feedback information, the comprehensive fatigue score and the movement success rate, generating a plan feedback.

[0027] Further, based on the preset standard movement key point sequence and the movement image data, the difference between the user's actual movement and the user's correct movement is quantified to obtain gesture difference data, including:

[0028] based on the preset key point positioning area, extracting the coordinates of the key points from the movement image data to obtain a real-time movement key point coordinate sequence;

[0029] based on the real-time movement key point coordinate sequence and the preset joint node set, calculating the angle change of the joint in the user's movement, obtaining gesture angle change data; the joint node set is used to represent the coordinates of the joint nodes participating in the action in the user's movement;

[0030] based on the gesture angle change data, extracting the periodic feature of the user's movement to obtain an actual movement period parameter;

[0031] based on the standard movement key point sequence and the actual movement period parameter, determining the key point coordinate sequence of the user's correct movement to obtain a standard reference key point coordinate sequence;

[0032] calculating the difference between the real-time movement key point coordinate sequence and the standard reference key point coordinate sequence to obtain gesture key point change data;

[0033] based on the gesture key point change data and the gesture angle change data, obtaining the gesture difference data.

[0034] Further, based on the gesture angle change data, the gesture key point change data and the standard movement key point sequence, generating auditory feedback information, including:

[0035] Based on the standard reference key point coordinate sequence and the posture angle change data, the angle change of the joint when the user moves is calculated from the angle change of the standard posture, and a joint angle deviation sequence is obtained;

[0036] Based on the preset deviation threshold, the joints and key points are extracted from the joint angle deviation sequence and the posture key point change data, and a deviation node sequence is obtained;

[0037] Based on the preset deviation weight, the correction order of the joints and key points in the deviation node sequence is determined, and a node feedback order sequence is obtained;

[0038] Based on the preset language database and the sorting rule, the node feedback order sequence is converted to obtain a language feedback text; and based on the language feedback text, auditory feedback information is generated.

[0039] Further, based on the fatigue image data and the fatigue voice data, the fatigue degree of the user is predicted to obtain a comprehensive fatigue score, including:

[0040] According to the fatigue image data, a preset convolutional neural network model is used to analyze the facial fatigue features of the user to obtain a facial fatigue score;

[0041] The mel-frequency cepstral coefficient feature of the fatigue voice data is extracted to obtain a mel-frequency cepstral coefficient feature vector;

[0042] Based on a preset time series analysis model, the time series features of the mel-frequency cepstral coefficient feature vector are analyzed to obtain a voice fatigue score; the time series analysis model is constructed based on a long short-term memory network;

[0043] Based on the attention mechanism, the facial fatigue score and the voice fatigue score are fused to obtain a comprehensive fatigue score.

[0044] Further, using the plan feedback and / or the historical training data of the user, the initial motion plan is updated to obtain an updated initial motion plan, including:

[0045] Based on the plan feedback and the historical training data, the effect of the initial motion plan is quantified to obtain a multi-target reward value; the multi-target reward value includes at least one of a motion effect reward value, a motion safety reward value and a motion comfort reward value;

[0046] Based on the multi-target reward value, the plan feedback, the historical training data and a preset policy gradient update model, a plan optimization parameter is obtained; the plan optimization parameter is used to indicate at least one aspect of adjusting the motion type, the motion frequency, the motion time and the motion heart rate interval of the user motion;

[0047] Based on preset safety constraints and plan optimization parameters, the initial exercise plan is updated to obtain the updated initial exercise plan.

[0048] Secondly, this application also provides an artificial intelligence-based rehabilitation training optimization system, including:

[0049] The basic data acquisition module is used to acquire the user's basic physiological data; the basic physiological data includes at least the cause of heart damage, age parameters, heart rate parameters, myocardial pressure parameters, and body mass index;

[0050] The initial plan creation module is used to generate an initial exercise plan for the user based on basic physiological data; the initial exercise plan is used to indicate the type of exercise, frequency, duration and heart rate zone of the user's exercise.

[0051] The real-time data acquisition module is used to acquire the user's motion image data, real-time heart rate parameters, fatigue image data, and fatigue voice data when executing the initial exercise plan;

[0052] The plan feedback generation module is used to quantify the effect of the initial exercise plan on the user based on motion image data, real-time heart rate parameters, fatigue image data, and fatigue voice data, and obtain the plan feedback of the initial exercise plan.

[0053] The initial plan update module is used to update the initial exercise plan using plan feedback and / or the user's historical training data, resulting in an updated initial exercise plan.

[0054] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement any of the artificial intelligence-based rehabilitation training optimization methods described in the first aspect of this application.

[0055] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements any of the artificial intelligence-based rehabilitation training optimization methods described in the first aspect of this application.

[0056] The aforementioned AI-based rehabilitation training optimization methods, systems, equipment, and media acquire basic physiological data from users. This data includes at least the cause of cardiac injury, age parameters, heart rate parameters, myocardial pressure parameters, and body mass index. Based on this data, an initial exercise plan is generated for the user. This plan specifies the type, frequency, duration, and heart rate zone of the exercise. During the execution of the initial exercise plan, the system acquires motion image data, real-time heart rate parameters, fatigue image data, and fatigue voice data. Based on these data, the system quantifies the impact of the initial exercise plan on the user, providing feedback. Using this feedback and / or the user's historical training data, the initial exercise plan is updated, resulting in a revised plan. This allows rehabilitation training to adapt to individual user characteristics and real-time changes in their condition, ensuring a safe and controllable training process and improving the personal adaptability of rehabilitation outcomes. Furthermore, the real-time feedback and dynamic adjustment mechanisms effectively enhance user compliance and comfort during training, thereby increasing the efficiency of cardiac rehabilitation training. Attached Figure Description

[0057] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0058] Figure 1 A schematic diagram illustrating the implementation environment of an artificial intelligence-based rehabilitation training optimization method according to one embodiment of this application;

[0059] Figure 2 A flowchart illustrating an artificial intelligence-based rehabilitation training optimization method provided in one embodiment of this application;

[0060] Figure 3 A schematic diagram of the structure of an artificial intelligence-based rehabilitation training optimization system provided in one embodiment of this application;

[0061] Figure 4 This is a schematic diagram of the structure of a computer device for an artificial intelligence-based rehabilitation training optimization method provided in one embodiment of this application. Detailed Implementation

[0062] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0063] The rehabilitation training optimization method based on artificial intelligence provided in this application embodiment describes the implementation environment of this application embodiment. Please refer to [link / reference]. Figure 1 The implementation environment includes a planning and monitoring terminal 101, a communication network 102, and a user training terminal 103. The planning and monitoring terminal 101 can be, but is not limited to, various personal computers, laptops, smartphones, tablets, IoT devices, and other electronic devices; the user training terminal 103 can be, but is not limited to, various augmented reality devices, smart bracelets, smartwatches, portable electrocardiographs, and other wearable devices. Both the planning and monitoring terminal 101 and the user training terminal 103 are equipped with programs that can receive and output information. The planning and monitoring terminal 101 can use this program to acquire the physiological information of the user receiving cardiac rehabilitation training and real-time training images, and output rehabilitation training plans and training action reminders. The user training terminal 103 can use this program to receive rehabilitation training plans and output the user's physiological information and real-time training images. The planning and monitoring terminal 101 communicates with the user training terminal 103 through the communication network 102, which can be a wireless network or a wired network.

[0064] Furthermore, the monitoring terminal 101 also has information storage and processing functions, which are used to formulate and adjust rehabilitation training plans based on the user's physiological information and real-time training images, and also to store the user's past training data.

[0065] In one exemplary embodiment, such as Figure 2 As shown, an artificial intelligence-based method for optimizing rehabilitation training is provided, which can be applied to... Figure 1 Taking the planning monitoring terminal as an example, the following S101-S105 are included:

[0066] S101, obtain the user's basic physiological data; the basic physiological data includes at least the cause of heart damage, age parameters, heart rate parameters, myocardial pressure parameters, and body mass index.

[0067] Specifically, the monitoring terminal will acquire basic physiological data from users receiving cardiac rehabilitation training. This basic physiological data includes at least the cause of cardiac injury, age parameters, heart rate parameters, myocardial pressure parameters, and body mass index (BMI). The cause of cardiac injury is the user's type of cardiac health condition, which can be obtained from doctor's diagnoses, medical records, or surgical reports. The age parameter is the user's actual age. The heart rate parameter is the user's resting heart rate, representing the number of heartbeats per minute at rest, and can be obtained based on the user's actual condition. The myocardial pressure parameter includes systolic and diastolic blood pressure values, representing the pressure load on the cardiovascular system. The BMI is a body mass index calculated based on height and weight.

[0068] S102 generates an initial exercise plan for the user based on basic physiological data; the initial exercise plan is used to indicate the type of exercise, frequency of exercise, duration of exercise, and heart rate zone of exercise for the user.

[0069] Specifically, the monitoring terminal determines a safe and effective exercise heart rate zone based on age and heart rate parameters using an exercise intensity calculation formula. This formula adjusts the estimated maximum heart rate based on age, and then combines the heart rate parameter with a preset intensity coefficient to define an appropriate heart rate training zone. The monitoring terminal uses a machine learning model to learn the relationship between the causes of heart injury, body mass index (BMI), and the ultimately selected exercise type in historical data, constructing an exercise type classification model. This model is then used to comprehensively analyze the user's causes of heart injury and BMI to determine the suitable exercise type for the user. The monitoring terminal also uses a regression analysis model to establish a correspondence between cardiac pressure parameters, BMI, and exercise frequency and duration, allowing it to estimate the appropriate exercise frequency and duration based on the user's cardiac pressure parameters and BMI. Finally, the monitoring terminal integrates the four parameters—exercise type, exercise frequency, exercise time, and exercise heart rate zone—obtained from the aforementioned calculations to generate the user's initial exercise plan.

[0070] S103, when executing the initial exercise plan, acquires the user's exercise image data, real-time heart rate parameters, fatigue image data, and fatigue voice data.

[0071] Specifically, when the user's training terminal executes the initial exercise plan, the plan monitoring terminal acquires the user's motion image data, real-time heart rate parameters, fatigue perception image data, and fatigue perception voice data. For example, the motion image data consists of continuous frames of the user performing training movements, used for subsequent analysis of the correctness of the exercise posture; each frame is timestamped. Optionally, the real-time heart rate parameter is the heart rate parameter continuously monitored during the user's exercise, reflecting the user's body's adaptation to the exercise intensity. For example, the fatigue perception image data is a close-up image of the user's facial expression, and the fatigue perception voice data is the user's voice information during exercise; the fatigue perception image data and fatigue perception voice data are used to infer the user's subjective feeling of fatigue.

[0072] S104 quantifies the effect of the initial exercise plan on the user based on motion image data, real-time heart rate parameters, fatigue image data, and fatigue voice data, and obtains the plan feedback of the initial exercise plan.

[0073] Specifically, the monitoring terminal determines the user's standard posture based on the initial exercise plan. By comparing the differences between the motion image data and the standard posture, it obtains a posture correctness score and auditory feedback information. The monitoring terminal also calculates whether the real-time heart rate falls within the exercise heart rate range of the initial exercise plan, obtaining a heart rate compliance score. Furthermore, it infers the user's fatigue sensation based on fatigue image data and fatigue voice data, obtaining a comprehensive fatigue score. Finally, the monitoring terminal comprehensively estimates the accuracy of the user's actions based on the aforementioned posture correctness score and heart rate compliance score, obtaining the exercise success rate. Finally, the monitoring terminal integrates the obtained exercise success rate, auditory feedback information, and comprehensive fatigue score to obtain the planning feedback for the initial exercise plan.

[0074] S105, using plan feedback and / or the user's historical training data, update the initial exercise plan to obtain the updated initial exercise plan.

[0075] Specifically, the planning monitoring terminal analyzes the current training plan feedback and historical training data based on the plan feedback and / or the user's historical training data, establishes a mapping relationship from training effect to plan parameter adjustment, identifies the parameters that need to be optimized among the four aspects of the initial exercise plan, including exercise type, exercise frequency, exercise time, and exercise heart rate zone, and obtains the optimized parameters to form an updated initial exercise plan.

[0076] This embodiment provides an AI-based rehabilitation training optimization method. It acquires basic physiological data from users and generates an initial exercise plan based on multi-parameter decision-making. During plan execution, it monitors the user's status in real time using multimodal data. By quantitatively analyzing the difference between actual exercise and the plan, it accurately and comprehensively evaluates the user's training effect, obtaining plan feedback. Furthermore, it dynamically adjusts the exercise plan based on plan feedback and historical data to achieve continuous optimization. This enables rehabilitation training to adapt to individual user characteristics and real-time status changes, ensuring the safety and controllability of the training process and improving the personal adaptability of rehabilitation effects. Simultaneously, the real-time feedback and dynamic adjustment mechanism effectively enhances user compliance and comfort during training, improving the efficiency of cardiac rehabilitation training.

[0077] In one embodiment, an initial exercise plan for the user is generated based on basic physiological data, including:

[0078] S201, based on preset exercise intensity parameters, combined with age and heart rate parameters, calculates exercise heart rate zone data using the following formula:

[0079]

[0080] in, It is the exercise heart rate zone. It's a heart rate parameter. It is the primary motion intensity parameter. It is an age parameter. It is the second motion intensity parameter.

[0081] Specifically, the monitoring terminal calculates the user's exercise heart rate zone data using a formula based on preset exercise intensity parameters, combined with age and heart rate parameters. For example, the exercise heart rate zone... It is a heart rate range value that includes upper and lower limits, used to indicate the range of heart rate activity during user exercise. Optionally, the heart rate parameter These are the heart rate parameter and age parameter from the aforementioned user basic physiological data. This refers to the age parameter from the aforementioned acquired user basic physiological data. For example, the first exercise intensity parameter... It is the recommended lower limit coefficient for intensity level in cardiac rehabilitation training, and the second exercise intensity parameter. This is the upper limit coefficient of the intensity level recommended in cardiac rehabilitation training, the primary exercise intensity parameter. Second motion intensity parameter The settings can be configured according to cardiac rehabilitation guidelines and the user's specific circumstances; the default settings are available. It is 0.5. It is 0.7.

[0082] S202, input the causes of heart injury and body mass index into the preset exercise type classification model to obtain exercise type data; the exercise type classification model is constructed based on the random forest algorithm.

[0083] Specifically, the monitoring terminal inputs the cause of cardiac injury and body mass index (BMI) into a preset exercise type classification model to obtain exercise type data. This exercise type classification model is constructed using a random forest algorithm. The monitoring terminal uses this random forest algorithm to construct the preset exercise type classification model based on the cause of cardiac injury, BMI, and selected exercise type from different historical patients' historical rehabilitation data. For example, the preset historical rehabilitation data can be obtained from actual work records; the cause of cardiac injury can be the patient's cardiac health status type, obtained from doctor's diagnoses, medical records, or surgical reports; the BMI is calculated based on the patient's height and weight; and the selected exercise type can be determined by a doctor or by the patient's own exercise choices.

[0084] S203 inputs cardiac pressure parameters and body mass index into a preset multiple linear regression model to obtain exercise frequency data and exercise time data.

[0085] Specifically, the monitoring terminal inputs cardiac pressure parameters and body mass index (BMI) into a preset multiple linear regression model to obtain exercise frequency and exercise time data. The preset multiple linear regression model is based on a multiple linear regression algorithm, learning the correlation patterns between cardiac pressure parameters, BMI, exercise frequency, and single exercise time among different historical patients in the preset historical exercise data. The multiple linear regression model is a statistical analysis method that makes predictions by establishing linear relationships between multiple independent variables and a dependent variable. For example, the preset historical exercise data can be obtained from actual work records. The cardiac pressure parameters are the systolic and diastolic blood pressure values ​​of historical patients, representing the pressure load on their cardiovascular system; the BMI is calculated based on the patient's height and weight; the exercise frequency specifies the number of training sessions per week; and the exercise time sets the duration of a single training session. The exercise frequency and duration that yielded the best training results for historical patients are recorded.

[0086] S204 generates an initial exercise plan based on exercise heart rate zone data, exercise type data, exercise frequency data, and exercise time data.

[0087] Specifically, the monitoring terminal integrates exercise heart rate zone data, exercise type data, exercise frequency data, and exercise time data to obtain an initial exercise plan.

[0088] This embodiment provides an AI-based method for optimizing rehabilitation training. It determines a safe and effective range of exercise intensity through formulaic calculation of physiological parameters, obtaining exercise heart rate zone data. Furthermore, it utilizes a machine learning model to select the most suitable exercise type based on the user's cardiac condition and body composition characteristics, obtaining exercise type data. Finally, it uses a regression analysis model to determine appropriate exercise frequency and duration based on cardiovascular status and weight indicators, obtaining exercise frequency and duration data. Integrating the exercise heart rate zone data, exercise type data, exercise frequency data, and exercise duration data yields an initial exercise plan. This method achieves a balance between medical safety standards and individualized needs, overcoming the subjectivity and limitations of traditional experience-based programs. It provides precise, scientific, safe, and reliable exercise guidance for cardiac rehabilitation patients, laying a solid foundation for subsequent training execution and dynamic optimization, and improving the efficiency of cardiac rehabilitation training.

[0089] In one embodiment, the effect of the initial exercise plan on the user is quantified based on motion image data, real-time heart rate parameters, fatigue image data, and fatigue voice data, resulting in plan feedback for the initial exercise plan, including:

[0090] S301, based on the preset standard motion key point sequence and motion image data, quantifies the difference between the user's actual movement and the user's correct movement to obtain posture difference data; posture difference data includes posture angle change data and posture key point change data; the standard motion key point sequence is used to characterize the user's standard posture during movement.

[0091] Specifically, the monitoring terminal extracts key point coordinates from motion image data during user training and obtains key point coordinates for standard postures from a standard motion key point sequence based on the type of movement in the initial exercise plan. By comparing the key point coordinates during user training with those for standard postures, the difference between the user's actual and correct movements is quantified, resulting in posture difference data that includes posture angle change data and posture key point change data. The standard motion key point sequence is a predefined set of spatial coordinate data representing standard movement postures, which can be obtained based on standard movement templates developed by professional rehabilitation physicians, providing a reference benchmark for movement postures. Posture angle change data is obtained by calculating the difference between joint angles during actual user movement and standard angles, reflecting the accuracy of joint range of motion during movement. Posture key point change data is obtained by comparing the differences between the user's body key point coordinates during movement and standard coordinates, representing the degree of deviation in the spatial position of various body parts. For example, key points and joint nodes can be set according to actual work requirements. Optionally, the preset standard motion key point sequence can take the form of… ,in It's sports type data. It is the standard posture cycle. These are the coordinates of key points in the standard pose. For example, both the pose angle change data and the pose key point change data are time-labeled to represent the user's training status at the timestamp corresponding to the timestamp in the motion image data.

[0092] S302, based on the change data of posture key points, quantifies the difference between the coordinates of key points during user movement and the coordinates of key points in standard posture, and obtains a posture correctness score.

[0093] Specifically, the planning monitoring terminal uses the following formula based on the posture key point change data: The difference between the coordinates of key points during user movement and the coordinates of key points in a standard posture is calculated to obtain the user training variance; the following formula is also used: The pose correctness score is calculated based on the difference in user training. In the aforementioned formula, It is a moment The degree of difference It is the total number of key points. It is a moment any key point The data on changes in key pose points, It is a moment The correctness score of posture This is a scaling parameter used to control the sensitivity of scores to differences, and can be set according to the actual work.

[0094] S303 generates auditory feedback information based on posture angle change data, posture key point change data, and standard motion key point sequence; the auditory feedback information is used to instruct the user to adjust the training posture through auditory information.

[0095] Specifically, the monitoring terminal plans to determine the deviation between the user's posture and the standard posture during real-time training based on posture angle change data, posture key point change data, and standard motion key point sequence, and generate auditory feedback information based on the deviation to instruct the user to adjust the training posture through auditory information.

[0096] S304 predicts the user's fatigue level and obtains a comprehensive fatigue score based on fatigue image data and fatigue voice data.

[0097] Specifically, the monitoring terminal obtains user facial expression image sequences through fatigue image data, extracts visual features including but not limited to the degree of eye closure and facial muscle relaxation, and quantifies the visual features to obtain a facial fatigue score; it also collects user voice signals through fatigue voice data, including acoustic features such as speech rate, pitch, and pronunciation clarity, and quantifies the acoustic features to obtain a voice fatigue score. The monitoring terminal combines the facial fatigue score and the voice fatigue score to obtain a comprehensive fatigue score.

[0098] The S305 quantifies the normality of a user's heart rate during exercise based on real-time heart rate parameters and an initial exercise plan, thus obtaining the heart rate compliance score.

[0099] Specifically, the monitoring terminal calculates the proportion of time during which the user's real-time heart rate parameters fall within the heart rate range required by the initial exercise plan, thus obtaining the heart rate compliance rate.

[0100] S306 uses a preset weighting system to weight and fuse the posture correctness score and heart rate compliance score to obtain the exercise success rate.

[0101] Specifically, the monitoring terminal plans to use the following formula based on preset weighting: The success rate of the exercise is obtained by weighting and fusing the posture correctness score and heart rate compliance score. In the aforementioned formula, It's the success rate of the exercise. It is the weight of the posture correctness score. It is a score for correct posture. It is the heart rate conformity weight. It refers to heart rate compliance. The preset weighting includes weights for posture correctness and heart rate compliance, which can be set according to actual work requirements.

[0102] S307 generates planning feedback based on auditory feedback information, comprehensive fatigue score, and exercise success rate.

[0103] Specifically, the planning monitoring terminal integrates auditory feedback information, comprehensive fatigue score, and exercise success rate to obtain planning feedback.

[0104] This embodiment provides an AI-based method for optimizing rehabilitation training. It quantifies the quality of user movement execution through precise postural difference analysis and generates specific auditory guidance to help users correct movement deviations in real time. Simultaneously, it accurately assesses user fatigue through multimodal information fusion, evaluates the appropriateness of training intensity based on heart rate monitoring data, and generates a comprehensive feedback plan reflecting the training effect by integrating various indicators. This achieves all-round monitoring and evaluation of the training process, provides timely corrective guidance, and significantly improves the safety, effectiveness, and user experience of rehabilitation training through the organic combination of precise quantitative assessment and real-time feedback guidance. It provides reliable technical support for personalized cardiac rehabilitation and improves the efficiency of cardiac rehabilitation training.

[0105] In one embodiment, based on a preset standard motion key point sequence and motion image data, the difference between the user's actual movement and the user's correct movement is quantified to obtain posture difference data, including:

[0106] S401, based on the preset key point positioning area, extracts the coordinates of key points from motion image data to obtain a real-time motion key point coordinate sequence.

[0107] Specifically, the monitoring terminal extracts the coordinates of key points from motion image data based on a preset key point localization area to obtain a real-time motion key point coordinate sequence. For example, the preset key point localization area refers to a predefined human joint detection region in the image, including the positions of major motion joints such as the shoulder, elbow, hip, and knee joints. This region can be set according to actual work requirements, and the key point information is identical to that of the preset standard motion key point sequence. Optionally, the monitoring terminal employs a deep learning-based key point detection algorithm, using a convolutional neural network to extract features and locate joints in the input image, outputting continuous key point coordinate data to obtain a real-time motion key point coordinate sequence. Optionally, the real-time motion key point coordinate sequence includes a time label to represent the user's training status at the corresponding time label in the motion image data.

[0108] S402 calculates the angle changes of joints during user movement based on the real-time motion key point coordinate sequence and the preset joint node set, and obtains posture angle change data; the joint node set is used to represent the coordinates of the joint nodes involved in the user's movement.

[0109] Specifically, the planning and monitoring terminal determines the coordinate data of key points belonging to preset joint nodes within the real-time motion key point coordinate sequence based on the real-time motion key point coordinate sequence and a preset set of joint nodes. It then calculates the angle values ​​by plotting the vector angles formed by adjacent key points based on the aforementioned preset joint coordinate data. These angle values ​​are then combined according to the time sequence of the real-time motion key point coordinate sequence to construct the angle change curves of each joint during movement, thus obtaining posture angle change data. The joint node set represents the coordinates of the joint nodes involved in the user's movement and can be set according to the actual task, such as hip, knee, and ankle. For example, the planning and monitoring terminal determines any timestamp in the real-time motion key point coordinate sequence based on the real-time motion key point coordinate sequence and the preset set of joint nodes. The coordinate data of key points of the hip in the preset joint nodes Coordinate data of key points on the knee Coordinate data of key points on the ankle. Using formula Obtain the first knee joint vector Using formula Obtain the second knee joint vector Then use the formula Get timestamp knee joint angle Knee joint angle within the timestamp of motion image data The knee joint angle change data obtained by differentiating with respect to time is a component of the posture angle change data. The number of joint angle change data included in the posture angle change data is determined by the number of joints included in the preset set of joint nodes.

[0110] S403 extracts the periodic features of user movement based on posture angle change data to obtain actual movement period parameters.

[0111] Specifically, the monitoring terminal analyzes the periodic changes in joint angles based on posture angle variation data using signal processing techniques to identify repetitive movement patterns. For example, gait cycles are determined by the periodic fluctuations in knee joint angles during walking. Optionally, dominant frequencies and cycle lengths can be detected from the angle variation sequence using autocorrelation analysis or Fourier transform methods.

[0112] S404, based on the standard motion key point sequence and the actual motion cycle parameters, determines the key point coordinate sequence for the user's correct motion, and obtains the standard reference key point coordinate sequence.

[0113] Specifically, the monitoring terminal determines the key point coordinate sequence for correct user movement from a standard motion key point sequence based on actual motion cycle parameters, thus obtaining a standard reference key point coordinate sequence. For example, the preset standard motion key point sequence can take the form of... ,in It's sports type data. It is the standard posture cycle. These are the key point coordinates for the standard posture. The planning and monitoring terminal determines the sequence of key point coordinates for correct user movement by matching the actual motion cycle parameters with the standard posture cycle.

[0114] S405 calculates the difference between the real-time motion keypoint coordinate sequence and the standard reference keypoint coordinate sequence to obtain the posture keypoint change data.

[0115] Specifically, the planning monitoring terminal calculates the Euclidean distance between each key point in the real-time motion key point coordinate sequence and the corresponding key point in the standard reference key point coordinate sequence to obtain the posture key point change data.

[0116] S406, based on the change data of pose key points and the change data of pose angles, obtain pose difference data.

[0117] Specifically, the planning monitoring terminal will combine the posture key point change data and posture angle change data to obtain posture difference data.

[0118] This embodiment provides an AI-based method for optimizing rehabilitation training. It extracts spatial information of human movement from visual data using keypoint detection technology to obtain a real-time sequence of keypoint coordinates. It quantifies the activity state of each joint during movement through joint angle calculation, obtaining posture angle change data. It identifies the rhythmic characteristics of individual movements through periodic analysis and achieves precise alignment with a standard template. Furthermore, it quantifies the deviation between actual and standard movements through difference calculation, obtaining posture keypoint change data. Finally, it integrates the posture keypoint change data and posture angle change data to generate posture difference data. This method achieves multi-dimensional and precise quantitative analysis of user movement postures, balancing spatial accuracy and temporal rhythm matching. It provides a reliable data foundation for real-time movement guidance and quality assessment. Through precise posture difference detection and detailed deviation analysis, it significantly improves the targeting and effectiveness of movement correction in rehabilitation training.

[0119] In one embodiment, auditory feedback information is generated based on posture angle change data, posture key point change data, and a standard motion key point sequence, including:

[0120] S501, based on the standard reference key point coordinate sequence and posture angle change data, calculates the difference between the angle change of the joints during user movement and the angle change of the standard posture, and obtains the joint angle deviation sequence.

[0121] Specifically, the monitoring terminal calculates the standard joint angle change data for the user's correct posture based on the standard reference keypoint coordinate sequence table, combined with the standard posture cycle and a preset set of joint nodes in the standard reference keypoint coordinate sequence. By calculating the difference between the standard joint angle change data and the posture angle change data for each joint, a joint angle deviation sequence is obtained. For example, the knee joint angle deviation is calculated as: actual knee angle in the posture angle change data - standard knee angle in the standard joint angle change data. Optionally, the data in the joint angle deviation sequence includes joint labels and time labels.

[0122] S502, based on a preset deviation threshold, extracts joints and key points from the joint angle deviation sequence and posture key point change data to obtain a deviation node sequence.

[0123] Specifically, the preset deviation thresholds include joint angle deviation thresholds and key point deviation thresholds. The monitoring terminal uses the joint angle deviation threshold to filter out key points of joints whose angle changes are greater than the joint angle deviation threshold from the joint angle deviation sequence; it also uses the key point deviation threshold to find key points with key point deviations greater than the key point deviation threshold from the posture key point change data, forming a deviation node sequence. For example, the preset deviation thresholds can be set according to the angle deviation requirements of each joint and the deviation requirements of key points in actual work, such as a knee joint angle deviation threshold of ±5 degrees and a hip joint angle deviation threshold of ±3 degrees.

[0124] S503, based on the preset deviation weight, determines the correction order of joints and key points in the deviation node sequence, and obtains the node feedback sequence.

[0125] Specifically, the planning and monitoring terminal determines the correction order of joints and key points in the deviation node sequence based on preset deviation weights, thus obtaining a node feedback sequence. For example, the preset deviation weights characterize the importance and safety priority of key point adjustments during movement, and can be set based on medical knowledge and principles of sports biomechanics. For instance, the knee joint is more critical than the elbow joint in supporting body weight, and therefore has a higher weight. The planning and monitoring terminal arranges the nodes in order of correction from urgent to normal, forming the node feedback sequence.

[0126] S504, based on a preset language database and sorting rules, transforms the node feedback sequence to obtain language feedback text; and generates auditory feedback information based on the language feedback text.

[0127] Specifically, the monitoring terminal plans to assemble the node feedback sequence into language feedback text according to a preset language database and sorting rules, and then generate auditory feedback information based on this language feedback text. For example, the language database is a pre-stored library of voice commands containing standard correction statement templates for various types of deviations; the sorting rules specify the organization of multiple feedback messages, such as arranging them in descending order of priority or merging similar items; these rules can be configured according to actual work requirements.

[0128] This embodiment provides an AI-based rehabilitation training optimization method. It quantifies the specific differences between user movements and standard postures through joint angle deviation calculation, obtaining a joint angle deviation sequence. Furthermore, it identifies key problem areas requiring correction through a threshold screening mechanism, obtaining a deviation node sequence. Combined with preset deviation weights, it determines the optimal feedback order, and finally generates auditory feedback information through language conversion technology. This achieves fully automated processing from motion data detection to voice guidance generation. Real-time voice guidance helps users correct movement deviations promptly, effectively improving the quality and efficiency of rehabilitation training while reducing the user's cognitive burden, significantly enhancing the targeting and effectiveness of movement correction in rehabilitation training.

[0129] In one embodiment, based on fatigue image data and fatigue voice data, the user's fatigue level is predicted to obtain a comprehensive fatigue score, including:

[0130] S601: Based on fatigue image data, a preset convolutional neural network model is used to analyze the user's facial fatigue characteristics and obtain a facial fatigue score.

[0131] Specifically, the monitoring terminal uses a pre-defined convolutional neural network model to analyze the user's facial fatigue characteristics and obtain a facial fatigue score. For example, the pre-defined convolutional neural network model can be constructed according to the needs of facial feature extraction in actual work, enabling multi-layer feature extraction of the input facial image, identifying fatigue-related visual features such as eyelid closure frequency, pupil size changes, and the degree of drooping of the corners of the mouth, and outputting a facial fatigue score between 0 and 1, with higher values ​​indicating more severe fatigue.

[0132] S602, extract the Mel frequency cepstral coefficient features from the fatigue-feeling speech data to obtain the Mel frequency cepstral coefficient feature vector.

[0133] Specifically, the monitoring terminal performs pre-emphasis, framing, and windowing processing on fatigue-related speech data to obtain pre-processed speech data. Then, it uses a Fast Fourier Transform (FFT) to convert the time-domain pre-processed speech data into a frequency-domain representation, obtaining frequency-domain speech data. Next, it utilizes a Mel-scale filter bank to simulate the frequency perception characteristics of the human ear, obtaining the Mel-frequency cepstral coefficient feature vector through logarithmic operations and discrete cosine transform. The Mel-frequency cepstral coefficient is a speech feature representation method based on human auditory characteristics, extracting speech signal features by simulating the human ear's sensitivity to different frequency sounds. For example, the Mel-frequency cepstral coefficient feature vector can reflect changes in speech characteristics caused by fatigue, such as slower speech rate, pitch changes, and unclear pronunciation.

[0134] S603, based on a preset temporal analysis model, analyzes the temporal characteristics of the Mel frequency cepstral coefficient feature vector to obtain a speech fatigue score; the temporal analysis model is constructed based on a long short-term memory network.

[0135] Specifically, the monitoring terminal inputs the Mel-frequency cepstral coefficient feature vector into a preset time-series analysis model to obtain a speech fatigue score. This time-series analysis model is constructed based on a Long Short-Term Memory (LSTM) network. An LTM network is a special type of recurrent neural network that, through the introduction of input, forget, and output gate mechanisms, can effectively learn and remember long-term dependencies in sequential data. For example, the preset time-series analysis model can be constructed using an LTM network based on characteristics such as the duration of fatigue-related speech data in actual work. This allows for the analysis of speech feature changes over time through the gating mechanism within the LTM network, capturing changes in dynamic speech characteristics caused by fatigue, such as the smoothness of intonation changes and the stability of speech rate, and outputting a speech fatigue score between 0 and 1 to reflect the degree of fatigue detected in the speech signal.

[0136] S604, based on an attention mechanism, fuses facial fatigue scores and voice fatigue scores to obtain a comprehensive fatigue score.

[0137] Specifically, the monitoring terminal plans to use an attention mechanism to weightedly fuse facial fatigue scores and voice fatigue scores to obtain a comprehensive fatigue score. The attention mechanism dynamically calculates the fusion weights of the two scores based on the signal-to-noise ratio, preset confidence level, and missing values ​​of the fatigue-related image and voice data. The facial fatigue score and voice fatigue score are then fused according to these weights to obtain the comprehensive fatigue score.

[0138] This embodiment provides an AI-based method for optimizing rehabilitation training. It uses a convolutional neural network to accurately extract visual fatigue features from facial images, obtaining a facial fatigue score; it captures acoustic fatigue features from speech signals using Mel-frequency cepstral coefficient analysis, obtaining a Mel-frequency cepstral coefficient feature vector; it models the temporal variation patterns of speech features using a long short-term memory network, obtaining a speech fatigue score; and it adaptively fuses the assessment results of visual and auditory modalities using an attention mechanism to obtain a comprehensive fatigue score. This method overcomes the limitations of single-modal assessment, comprehensively and accurately reflecting the user's true fatigue state, and significantly improves the intelligence level and security of the rehabilitation training system.

[0139] In one embodiment, the initial exercise plan is updated using plan feedback and / or the user's historical training data to obtain an updated initial exercise plan, including:

[0140] S701 quantifies the effect of initializing the exercise plan based on plan feedback and historical training data to obtain multi-objective reward values; the multi-objective reward values ​​include at least one of exercise effect reward value, exercise safety reward value and exercise comfort reward value.

[0141] Specifically, the planning and monitoring terminal calculates a multi-objective reward value based on auditory feedback information, overall fatigue score, and exercise success rate from the planning feedback, combined with historical training planning feedback included in historical training data. This multi-objective reward value comprises exercise effect reward value, exercise safety reward value, and exercise comfort reward value. For example, the exercise effect reward value is calculated using the following formula: ,in It is the reward value for exercise performance; It is the first weight, used to calculate the impact of the difference between the actual and target success rates on the exercise performance reward value. The second weight is used to calculate the impact of the difference between the success rate of the current exercise and the success rate of the previous exercise on the exercise effect reward value. The first and second weights can be set according to the actual work, and their sum is 1. It's the success rate of the exercise; It refers to the success rate of the target exercise, and training requirements can be set according to actual work needs. This is the difference between the current training session and the previous training session, which can be obtained from historical training data. Optionally, the exercise safety reward value can be calculated using the following formula: , It is a sports safety reward value. It's heart rate compliance. It is a comprehensive fatigue score. It is the fatigue threshold. It is the third weight. This is the fourth weight. For example, the exercise comfort reward value is calculated using the following formula: , It is the exercise comfort reward value. It is the fifth weight. It is a comprehensive fatigue score. It is the sixth weight. This is the duration of this training session, which can be obtained from the initial exercise plan. This is the historical average training time, which can be obtained from historical training data. Optionally, historical training data can be obtained from the user's historical training records.

[0142] S702, based on multi-objective reward values, plan feedback, historical training data and a preset policy gradient update model, obtains plan optimization parameters; the plan optimization parameters are used to indicate at least one aspect of adjusting the user's exercise type, exercise frequency, exercise time and exercise heart rate zone.

[0143] Specifically, the planning monitoring terminal inputs planning feedback and historical training data into a pre-defined policy gradient update model to obtain planning optimization parameters. This model uses multi-objective reward values ​​as the optimization objective and planning feedback and historical training data as state inputs. It calculates the gradient direction of the policy function and updates the model parameters. Based on the learned optimal mapping relationship from user state to planned adjustment actions, it obtains the planning optimization parameters. These parameters are specific adjustment suggestions output by the model, including indications of changes in exercise type, adjustment magnitudes of exercise frequency, increases or decreases in exercise time, and changes in the upper and lower limits of the exercise heart rate zone. These parameters provide specific operational guidance for subsequent plan updates. The pre-defined policy gradient update model can be constructed using a policy gradient update algorithm based on historical training data.

[0144] S703 updates the initial exercise plan based on preset safety constraint rules and plan optimization parameters, and obtains the updated initial exercise plan.

[0145] Specifically, the planning monitoring terminal filters the plan optimization parameters using preset safety constraint rules, adjusts parameters that do not conform to these rules, and updates the initial exercise plan using the adjusted optimization parameters, resulting in an updated initial exercise plan. For example, the preset safety constraint rules limit the safe range of plan optimization parameter adjustments and can be set according to cardiac rehabilitation medicine standards, including but not limited to the safe range of exercise heart rate, the maximum duration of a single training session, and exercise types corresponding to different types of heart injury. The planning monitoring terminal maps the plan optimization parameters to specific plan parameters that conform to the safety rules. For instance, if the heart rate range in the plan optimization parameters exceeds the safe range, it is truncated to the maximum or minimum allowed value; when the suggested exercise type does not match the user's cardiac condition, it is replaced with a safe alternative exercise type.

[0146] This embodiment provides an AI-based rehabilitation training optimization method. It comprehensively measures the overall performance of the training plan in terms of effectiveness, safety, and comfort through multi-dimensional quantitative evaluation, obtaining multi-objective reward values. Furthermore, it intelligently generates optimization directions based on historical data and real-time feedback using a reinforcement learning algorithm, obtaining optimization parameters. Finally, it employs preset safety constraints to ensure all optimization adjustments remain within medically permissible limits, updating the initial exercise plan to obtain an updated initial exercise plan. This achieves continuous self-optimization and personalized adaptation of the training plan. Safety constraints ensure controllable risks during the training process, allowing rehabilitation training to dynamically adjust according to changes in the user's condition. This avoids the limitations of traditional static plans that cannot adapt to individual progress and prevents potential safety hazards from over-optimization, significantly improving the targeting and effectiveness of movement correction in rehabilitation training.

[0147] The aforementioned AI-based rehabilitation training optimization methods, systems, equipment, and media acquire the user's basic physiological data. This basic physiological data includes at least the cause of cardiac injury, age parameters, heart rate parameters, myocardial pressure parameters, and body mass index. Based on this basic physiological data, an initial exercise plan is generated for the user. This initial exercise plan indicates the type, frequency, duration, and heart rate zone of the exercise. During the execution of the initial exercise plan, the user's motion image data, real-time heart rate parameters, fatigue image data, and fatigue voice data are acquired. Based on these data, the effect of the initial exercise plan on the user is quantified, resulting in feedback. Using this feedback and / or the user's historical training data, the initial exercise plan is updated, resulting in an updated initial exercise plan. This enables rehabilitation training to adapt to individual user characteristics and real-time changes in their condition, ensuring the safety and controllability of the training process and improving the personal adaptability of rehabilitation effects. Furthermore, the real-time feedback and dynamic adjustment mechanism effectively enhances user compliance and comfort during training, thereby improving the efficiency of cardiac rehabilitation training.

[0148] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0149] Based on the same inventive concept, this application also provides an AI-based rehabilitation training optimization system for implementing the AI-based rehabilitation training optimization method described above. The solution provided by this system is similar to the implementation scheme described in the above method. Therefore, the specific limitations of one or more AI-based rehabilitation training optimization system embodiments provided below can be found in the limitations of the AI-based rehabilitation training optimization method described above, and will not be repeated here.

[0150] In one exemplary embodiment, such as Figure 3 As shown, an artificial intelligence-based rehabilitation training optimization system 200 is provided, including:

[0151] The basic data acquisition module 201 is used to acquire the user's basic physiological data; the basic physiological data includes at least the cause of heart damage, age parameters, heart rate parameters, myocardial pressure parameters, and body mass index;

[0152] The initial plan formulation module 202 is used to generate an initial exercise plan for the user based on basic physiological data; the initial exercise plan is used to indicate the type of exercise, frequency of exercise, duration of exercise, and heart rate zone of exercise for the user.

[0153] The real-time data acquisition module 203 is used to acquire the user's motion image data, real-time heart rate parameters, fatigue image data, and fatigue voice data when executing the initial exercise plan;

[0154] The plan feedback generation module 204 is used to quantify the effect of the initial exercise plan on the user based on motion image data, real-time heart rate parameters, fatigue image data and fatigue voice data, and obtain the plan feedback of the initial exercise plan.

[0155] The initial plan update module 205 is used to update the initial exercise plan using plan feedback and / or the user's historical training data, so as to obtain the updated initial exercise plan.

[0156] Furthermore, the initial planning module is also used for:

[0157] Based on preset exercise intensity parameters, combined with age and heart rate parameters, the following formula is used to calculate the exercise heart rate zone data:

[0158]

[0159] in, It is the exercise heart rate zone. It's a heart rate parameter. It is the primary motion intensity parameter. It is an age parameter. It is the second motion intensity parameter;

[0160] The causes of heart injury and body mass index are input into a pre-defined exercise type classification model to obtain exercise type data; the exercise type classification model is constructed based on the random forest algorithm.

[0161] The cardiac pressure parameters and body mass index are input into a preset multiple linear regression model to obtain exercise frequency data and exercise time data.

[0162] An initial exercise plan is obtained based on exercise heart rate zone data, exercise type data, exercise frequency data, and exercise time data.

[0163] Furthermore, the plan feedback generation module includes:

[0164] The posture difference calculation unit is used to quantify the difference between the user's actual movement and the user's correct movement based on a preset standard motion key point sequence and motion image data, and to obtain posture difference data. The posture difference data includes posture angle change data and posture key point change data. The standard motion key point sequence is used to characterize the user's standard posture during movement.

[0165] The posture difference quantification unit is used to quantify the difference between the coordinates of key points during user movement and the coordinates of key points in standard posture based on posture key point change data, and to obtain a posture correctness score.

[0166] The auditory information generation unit is used to generate auditory feedback information based on posture angle change data, posture key point change data, and standard motion key point sequence; the auditory feedback information is used to instruct the user to adjust the training posture.

[0167] The comprehensive fatigue calculation unit is used to predict the user's fatigue level and obtain a comprehensive fatigue score based on fatigue image data and fatigue voice data.

[0168] The heart rate normality quantification unit is used to quantify the normality of the user's heart rate during exercise based on real-time heart rate parameters and an initial exercise plan, and to obtain the heart rate compliance.

[0169] The exercise comprehensive calculation unit is used to weight and fuse the posture correctness score and heart rate compliance according to preset weights to obtain the exercise success rate;

[0170] The planning feedback generation unit is used to generate planning feedback based on auditory feedback information, comprehensive fatigue score, and exercise success rate.

[0171] Furthermore, the pose difference calculation unit is also used for:

[0172] Based on the preset key point localization area, the coordinates of key points are extracted from the motion image data to obtain the real-time motion key point coordinate sequence;

[0173] Based on the real-time motion key point coordinate sequence and the preset joint node set, the angle changes of the joints during user movement are calculated to obtain posture angle change data; the joint node set is used to represent the coordinates of the joint nodes involved in the user movement.

[0174] Based on posture angle change data, the periodic features of user movement are extracted to obtain actual movement period parameters;

[0175] Based on the standard motion key point sequence and actual motion cycle parameters, the key point coordinate sequence for correct user motion is determined, and the standard reference key point coordinate sequence is obtained.

[0176] The difference between the real-time motion keypoint coordinate sequence and the standard reference keypoint coordinate sequence is calculated to obtain the posture keypoint change data;

[0177] Furthermore, the auditory information generation unit is also used for:

[0178] Based on the standard reference key point coordinate sequence and posture angle change data, the difference between the angle change of the joints during user movement and the angle change of the standard posture is calculated to obtain the joint angle deviation sequence.

[0179] Based on a preset deviation threshold, joints and key points are extracted from joint angle deviation sequence and posture key point change data to obtain deviation node sequence;

[0180] Based on the preset deviation weights, the correction order of joints and key points in the deviation node sequence is determined, and the node feedback sequence is obtained.

[0181] Based on a preset language database and sorting rules, the node feedback sequence is transformed to obtain language feedback text; and auditory feedback information is generated based on the language feedback text.

[0182] Furthermore, the integrated fatigue calculation unit is also used for:

[0183] Based on fatigue image data, a pre-set convolutional neural network model is used to analyze the user's facial fatigue characteristics and obtain a facial fatigue score;

[0184] Mel frequency cepstral coefficient features are extracted from fatigue-related speech data to obtain Mel frequency cepstral coefficient feature vectors;

[0185] Based on a pre-defined temporal analysis model, the temporal characteristics of the Mel frequency cepstral coefficient feature vector are analyzed to obtain the speech fatigue score; the temporal analysis model is constructed based on a long short-term memory network.

[0186] Based on the attention mechanism, facial fatigue score and voice fatigue score are fused to obtain a comprehensive fatigue score.

[0187] Furthermore, the initial plan update module is also used for:

[0188] Based on plan feedback and historical training data, the effect of the initial exercise plan is quantified to obtain multi-objective reward values; the multi-objective reward values ​​include at least one of exercise effect reward value, exercise safety reward value, and exercise comfort reward value;

[0189] Based on multi-objective reward values, plan feedback, historical training data, and a preset policy gradient update model, plan optimization parameters are obtained. These parameters are used to indicate adjustments to at least one aspect of the user's exercise: exercise type, exercise frequency, exercise time, and exercise heart rate zone.

[0190] Based on preset safety constraints and plan optimization parameters, the initial exercise plan is updated to obtain the updated initial exercise plan.

[0191] Based on the change data of pose key points and the change data of pose angles, pose difference data is obtained.

[0192] In one embodiment, such as Figure 4 A computer device is provided, comprising:

[0193] At least one processor 301, and a memory 302 communicatively connected to at least one of the processors 301: the memory stores application code executable by at least one of the processors, the application code being executed by at least one of the processors to enable at least one of the processors to perform the AI-based rehabilitation training optimization method as described above.

[0194] Computer equipment may also include: sensor 303.

[0195] The processor 301, memory 302 and sensor 303 can be connected via a bus or other means, with the bus being an example in the figure.

[0196] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps in the above method embodiments.

[0197] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The components described as separate parts may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this disclosure according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0198] The above-described embodiments are merely illustrative of several implementation methods of the embodiments of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the embodiments of this application, and these modifications and improvements all fall within the protection scope of the embodiments of this application.

Claims

1. An artificial intelligence-based rehabilitation training optimization method, characterized in that, The method comprises: obtaining basic physiological data of a user; the basic physiological data at least comprises a heart damage cause, an age parameter, a heart rate parameter, a heart pressure parameter and a body mass index; generating an initial exercise plan of the user based on the basic physiological data; the initial exercise plan is used to indicate an exercise type, an exercise frequency, an exercise time and an exercise heart rate interval of the user's exercise; obtaining exercise image data, real-time heart rate parameters, fatigue image data and fatigue voice data of the user when the initial exercise plan is executed; quantifying the effect of the initial exercise plan on the user based on the exercise image data, the real-time heart rate parameters, the fatigue image data and the fatigue voice data to obtain plan feedback of the initial exercise plan; updating the initial exercise plan using the plan feedback and / or historical training data of the user to obtain an updated initial exercise plan.

2. The method of claim 1, wherein, The generating of the initial exercise plan of the user based on the basic physiological data comprises: based on a preset exercise intensity parameter, combining the age parameter and the heart rate parameter, and using the following formula to calculate exercise heart rate interval data: wherein is an exercise heart rate interval, is a heart rate parameter, is a first exercise intensity parameter, is an age parameter, is a second exercise intensity parameter; inputting the heart damage cause and the body mass index into a preset exercise type classification model to obtain exercise type data; the exercise type classification model is constructed based on a random forest algorithm; inputting the heart pressure parameter and the body mass index into a preset multiple linear regression model to obtain exercise frequency data and exercise time data; based on the exercise heart rate interval data, the exercise type data, the exercise frequency data and the exercise time data, the initial exercise plan is obtained.

3. The method of claim 1, wherein, The quantifying of the effect of the initial exercise plan on the user based on the exercise image data, the real-time heart rate parameters, the fatigue image data and the fatigue voice data to obtain the plan feedback of the initial exercise plan comprises: based on a preset standard exercise key point sequence and the exercise image data, quantifying the difference between the user's actual exercise and the user's correct exercise to obtain posture difference data; the posture difference data comprises posture angle change data and posture key point change data; the standard exercise key point sequence is used to represent the standard posture of the user's exercise; based on the posture key point change data, quantifying the difference between the coordinates of the key points of the user's exercise and the coordinates of the key points of the standard posture to obtain a posture correctness score; based on the posture angle change data, the posture key point change data and the standard exercise key point sequence, generating auditory feedback information; the auditory feedback information is used to instruct the user to adjust the training posture through auditory information; based on the fatigue image data and the fatigue voice data, predicting the fatigue degree of the user to obtain a comprehensive fatigue score; based on the real-time heart rate parameters and the initial exercise plan, quantifying the normal degree of the user's heart rate when exercising to obtain a heart rate compliance degree; According to the preset weighting weight, the posture correctness score and the heart rate compliance are weighted and fused to obtain a movement success rate; Based on the auditory feedback information, the comprehensive fatigue score and the movement success rate, the plan feedback is generated.

4. The method of claim 3, wherein, Based on the preset standard motion key point sequence and the motion image data, the difference between the actual movement of the user and the correct movement of the user is quantified to obtain posture difference data, including: Based on the preset key point positioning area, the coordinates of the key points are extracted from the motion image data to obtain a real-time motion key point coordinate sequence; Based on the real-time motion key point coordinate sequence and a preset joint node set, the angle change of the joint when the user moves is calculated to obtain posture angle change data; the joint node set is used to represent the coordinates of the joint node participating in the action when the user moves; Based on the posture angle change data, the periodic feature of the movement of the user is extracted to obtain an actual movement period parameter; Based on the standard motion key point sequence and the actual movement period parameter, the key point coordinate sequence of the correct movement of the user is determined to obtain a standard reference key point coordinate sequence; The difference between the real-time motion key point coordinate sequence and the standard reference key point coordinate sequence is calculated to obtain the posture key point change data; Based on the posture key point change data and the posture angle change data, the posture difference data is obtained.

5. The method of claim 4, wherein, Based on the posture angle change data, the posture key point change data and the standard motion key point sequence, auditory feedback information is generated, including: Based on the standard reference key point coordinate sequence and the posture angle change data, the difference between the angle change of the joint when the user moves and the angle change of the standard posture is calculated to obtain a joint angle deviation sequence; Based on a preset deviation threshold, the joint and the key point are extracted from the joint angle deviation sequence and the posture key point change data to obtain a deviation node sequence; Based on a preset deviation weight, the correction order of the joint and the key point in the deviation node sequence is determined to obtain a node feedback order sequence; Based on a preset language database and a sorting rule, the node feedback order sequence is converted to obtain a language feedback text; and the auditory feedback information is generated based on the language feedback text.

6. The method of claim 3, wherein, Based on the fatigue image data and the fatigue voice data, the fatigue degree of the user is predicted to obtain a comprehensive fatigue score, including: According to the fatigue image data, a preset convolutional neural network model is used to analyze the facial fatigue features of the user to obtain a facial fatigue score; The mel-frequency cepstral coefficient feature of the fatigue voice data is extracted to obtain a mel-frequency cepstral coefficient feature vector; Based on a preset time sequence analysis model, the time sequence features of the mel-frequency cepstral coefficient feature vector are analyzed to obtain a voice fatigue score; the time sequence analysis model is constructed based on a long short-term memory network; Based on an attention mechanism, the facial fatigue score and the voice fatigue score are fused to obtain the comprehensive fatigue score.

7. The method of claim 3, wherein, The using the plan feedback and / or the historical training data of the user to update the initial exercise plan to obtain an updated initial exercise plan comprises: quantifying an effect of the initial exercise plan based on the plan feedback and the historical training data to obtain a multi-objective reward value; the multi-objective reward value comprises at least one of an exercise effect reward value, an exercise safety reward value, and an exercise comfort reward value; obtaining a plan optimization parameter based on the multi-objective reward value, the plan feedback, the historical training data, and a preset policy gradient update model; the plan optimization parameter is used to indicate an adjustment of at least one aspect of the exercise type, the exercise frequency, the exercise time, and the exercise heart rate interval of the user exercise; updating the initial exercise plan based on a preset safety constraint rule and the plan optimization parameter to obtain an updated initial exercise plan.

8. An artificial intelligence-based rehabilitation training optimization system, characterized in that, The system comprises: a basic data acquisition module configured to acquire basic physiological data of a user; the basic physiological data comprises at least a heart damage cause, an age parameter, a heart rate parameter, a heart pressure parameter, and a body mass index; an initial plan formulation module configured to generate an initial exercise plan of the user based on the basic physiological data; the initial exercise plan is used to indicate an exercise type, an exercise frequency, an exercise time, and an exercise heart rate interval of the user exercise; a real-time data acquisition module configured to acquire exercise image data, real-time heart rate parameters, fatigue image data, and fatigue voice data of the user when the initial exercise plan is executed; a plan feedback generation module configured to quantify an effect of the initial exercise plan on the user based on the exercise image data, the real-time heart rate parameters, the fatigue image data, and the fatigue voice data to obtain a plan feedback of the initial exercise plan; an initial plan update module configured to use the plan feedback and / or historical training data of the user to update the initial exercise plan to obtain an updated initial exercise plan. 9.A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the computer device is configured to perform the method according to any one of claims 1-8 when the computer program is executed by the processor. The processor executes the computer program to implement the steps of the method in any one of claims 1 to 7.

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

Images