An artificial intelligence-based pediatric massage teaching auxiliary system

Abstract

The present invention discloses an artificial intelligence-based pediatric massage teaching auxiliary system, which specifically relates to the technical field of massage auxiliary teaching. The system adopts Kalman filtering technology to effectively filter out large-scale interference in hand motion data, combines fast Fourier transform or wavelet transform to perform time-frequency analysis, and accurately captures the frequency domain characteristics of micro-tremor signals. Through fine micro-tremor signal analysis, noise interference is removed, and subtle changes in trainees' hand micro-tremors are accurately identified, which significantly improves the assessment accuracy of massage action stability. An integrated learning progress management module automatically adjusts teaching content according to the trainees' learning progress and skill mastery, provides real-time feedback and positive incentives, provides incentives when trainees perform well, and provides real-time prompts and optimizes subsequent actions when deviations occur, so as to help trainees continuously improve their techniques, avoid error accumulation, and effectively improve teaching effects.

Description

Technical Field

[0001] The present invention relates to the technical field of massage auxiliary teaching, and more specifically, to a pediatric massage teaching auxiliary system based on artificial intelligence. Background Art

[0002] As the application of traditional Chinese massage in pediatric health care and disease prevention becomes more and more popular, the requirements for precise control of massage techniques and operational stability are increasing. Traditional pediatric massage teaching mainly relies on experience transfer and master-apprentice demonstration, lacking objective quantitative means of subtle dynamic changes during the technique. In recent years, sensor technology and artificial intelligence algorithms have made significant progress in the field of motion analysis, providing new possibilities for real-time data acquisition, signal processing and abnormal state detection.

[0003] In the existing technology, there are problems such as asynchronous timing of multi-sensor data acquisition, large signal noise, and obvious physiological differences between individuals, which makes the accurate capture and processing of hand tremor data during massage face severe challenges. Specifically, how to effectively separate the main movement and tremor signals through advanced signal processing technology, and how to establish a normal tremor baseline based on standard massage data and dynamically adjust the abnormal scoring threshold using adaptive online learning are key technical problems that need to be solved. The solution to these problems will directly affect the accurate construction of the tremor feature vector and the robustness of anomaly detection, and thus determine the actual application effect and promotion value of the system in assisting pediatric massage teaching. Summary of the invention

[0004] In order to overcome the above-mentioned defects of the prior art, the present invention provides a pediatric massage teaching auxiliary system based on artificial intelligence to solve the problems raised in the above-mentioned background technology.

[0005] To achieve the above-mentioned purpose, the present invention provides the following technical solution: an artificial intelligence-based pediatric massage teaching auxiliary system, comprising:

[0006] Acquire hand motion data during the massage process through sensors, including hand tremor data during the massage process; analyze the hand tremor data to generate a feature vector including tremor amplitude characteristics, frequency characteristics, and tremor characteristics;

[0007] By learning the standard massage data, the abnormal scoring threshold is output; an adaptive online learning algorithm is introduced to establish a dynamic individualized normal microtremor baseline, and the abnormal scoring threshold is adjusted using a clustering algorithm;

[0008] The feature vector is analyzed based on the machine learning model, and an abnormal score is output. The abnormal score is used to determine whether it belongs to the abnormal microtremor type; when the abnormal score exceeds the abnormal score threshold, it is determined to be an abnormal microtremor type;

[0009] If the tremor is abnormal, prompt the student to adjust the gesture, relax the muscles, and optimize the angle of force; otherwise, no action will be taken.

[0010] Preferably, based on Kalman filtering technology, the interference of large-amplitude movements in hand motion data is removed, the tremor signal is retained, the tremor signal with the main motion signal removed is subjected to frequency domain analysis, the fast Fourier transform or wavelet transform time-frequency analysis method is applied to perform frequency division processing on the tremor signal, the tremor component within a specific frequency band is captured, and the micro-jitter feature is output; a feature vector is constructed based on the micro-jitter feature, and the feature vector includes the energy density distribution of the micro-tremor, the deviation of the main motion direction, and the time series correlation parameter.

[0011] Preferably, the hand motion data to be processed is obtained, low-pass / band-pass filtering and multi-dimensional time-frequency analysis are performed, improved adaptive filtering is implemented based on real-time spectrum data, the filtered signal and residual information are output, the multi-scale reconstruction technology is used to restore the time domain details of the micro-tremor and quantify the residual, the sliding window is used to extract multi-dimensional statistical features and weighted fusion is performed, the abnormal micro-tremor pattern is determined by comparison with standard massage data, and the filter parameters are optimized through online feedback closed-loop.

[0012] Preferably, a massage technique stability optimization module is included, which inputs the acquired feature vector into a stability evaluation model, outputs a manipulation stability index, and automatically generates real-time feedback and optimization suggestions based on the manipulation stability index; through a visual interface, the trainee can view his current manipulation stability index in real time; the stability evaluation model satisfies the following formula:

[0013] ;

[0014] Among them, S represents the technique stability index, M represents the total sampling time, and j represents the sampling sequence number. is the tremor amplitude at time j, is the standard reference amplitude, ≠0, represents the time-frequency energy density, is the angle between the micro-tremor direction and the main motion direction, and a and β are nonlinear adjustment factors.

[0015] Preferably, the system comprises:

[0016] The massage plan deviation optimization module is used to collect the strength, rhythm, angle, time, and trajectory data of the massage in real time, and compare the collected data with the preset personalized massage plan to output the massage plan execution quality index;

[0017] Automatically generate real-time feedback and optimization suggestions based on the execution quality index of the massage program; through the visual interface, students can view their current execution quality index of the massage program in real time; according to the manipulation stability index, students are prompted to adjust their gestures, relax their muscles, and optimize the angle of force in real time;

[0018] The method for obtaining the massage program execution quality index is as follows:

[0019] Collect and preprocess the key parameters and trajectory data of massage at each moment; extract the key points of the actual massage process and the preset plan, and calculate the matching degree of each key point in time and space; calculate the key parameter matching degree of the key parameter data in the sampling frame through a non-traditional geometric mean and exponential mapping method; based on the key point matching degree and the key parameter matching degree, jointly calculate the massage plan execution quality index through the harmonic mean method; this index can objectively reflect the deviation between the actual massage action and the personalized plan, and provide a quantitative basis for subsequent feedback and skill improvement.

[0020] Preferably, the method for obtaining the complete index of the massage program execution includes:

[0021] The key points of the trajectory curve of the execution process of the massage program and the trajectory curve of the personalized massage program are obtained respectively, wherein the key points refer to the turning points, the mutation points of the strength and the rhythm of the trajectory curve; the matching degree of the key points is calculated based on the time-space coordinates of the key points;

[0022] The key points (including trajectory inflection points, strength and rhythm mutation points) are extracted from the preset personalized massage program trajectory and the actual execution trajectory respectively. The preset and actual key point sequences are recorded as and , where each key point ; Represent the horizontal axis, vertical axis and time coordinate respectively;

[0023] Assume there are N key points, i represents the sequence number, and the key point matching degree is calculated by the following formula: :

[0024] ;

[0025] ;

[0026] Among them, di represents the Euclidean distance between the i-th key points;

[0027] Record the strength, rhythm, angle, time, and trajectory data of each massage plan, divide the execution process of the massage plan into several moments according to the number of sampling frames, and record the average value of strength, rhythm, angle, and time at each moment;

[0028] It represents the matching degree of the pth dimension, which includes strength, rhythm, angle, time, and trajectory data, that is, ;use represents the deviation of the pth dimension; suppose there are M sampling moments, j represents the sequence number, and the average value of intensity, rhythm, angle and time is recorded as the actual parameter sequence , the corresponding preset parameter sequence is recorded as ,and ;

[0029] The matching degree of key massage parameters is calculated by the following formula:

[0030] ;

[0031] ;

[0032] The matching degree of joint key points and the matching degree of massage key parameters are calculated by the formula

[0033] ;

[0034] The massage program execution quality index was calculated.

[0035] Preferably, the system comprises:

[0036] The learning progress management module obtains the performance of each student in the learning progress, obtains the massage plan execution quality index and the technique stability index, performs linear normalization on the massage plan execution quality index and the technique stability index, and calculates the learning progress evaluation coefficient by weighted summation with the task completion efficiency. The massage teaching content is adjusted based on the learning progress evaluation coefficient. When the learning progress evaluation coefficient is higher than the preset target value, it is considered that the current target massage task is executed well, and positive incentive information is fed back and the next stage target massage task is added.

[0037] Preferably, the system comprises:

[0038] A personalized massage program generation module, which is used to obtain a preset personalized massage program;

[0039] The personalized massage plan generation module is used to obtain children's physiological data and historical efficacy data, build a massage plan generation model based on a deep learning model, and train it through standardized historical data to learn the relationship between children's physiological data and massage efficacy; collect children's information in real time and input it into the trained model to generate a personalized massage plan; the personalized massage plan includes massage techniques, strength, rhythm, time, and trajectory parameters.

[0040] Technical effects and advantages of the present invention:

[0041] (1) The pediatric massage teaching auxiliary system provided by the present invention can dynamically generate massage plans and optimization suggestions suitable for students in real time by accurately evaluating the students' hand tremor signals and movement performances, thereby solving the problem that traditional teaching cannot provide personalized guidance based on individual differences. The Kalman filter technology is used to effectively filter out large-scale interference in hand motion data, and time-frequency analysis is performed in combination with fast Fourier transform (FFT) or wavelet transform to accurately capture the frequency domain characteristics of the tremor signal. Through fine tremor signal analysis, the system can remove the influence of noise and accurately identify subtle changes in the students' hand tremors, thereby improving the evaluation accuracy of the stability of the massage movements and solving the noise interference problem existing in traditional signal processing.

[0042] (2) The pediatric massage teaching auxiliary system provided by the present invention integrates a learning progress management module, automatically adjusts the massage teaching content according to the student's learning progress and skill mastery, and provides real-time positive incentives and feedback. When the student performs well during the massage process, the system will provide incentive information to enhance the student's learning motivation; when deviations are found, the system will prompt and optimize subsequent actions in real time, effectively avoiding the accumulation and transmission of errors, helping students to continuously improve their techniques, and solving the problem of lack of real-time progress monitoring and adjustment in traditional teaching. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 This is a structural block diagram of the pediatric massage teaching auxiliary system based on microtremor data of the present invention.

[0044] Figure 2 This is a structural block diagram of the massage teaching auxiliary system based on learning progress management of the present invention. DETAILED DESCRIPTION

[0045] The exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0046] At the same time, it should be understood that for the convenience of description, the sizes of the various parts shown in the drawings are not drawn according to the actual proportional relationship.

[0047] The following description of at least one exemplary embodiment is merely illustrative in nature and is in no way intended to limit the present application, its application, or uses.

[0048] Technologies, methods, and equipment known to ordinary technicians in the relevant art may not be discussed in detail, but where appropriate, the technologies, methods, and equipment should be considered as part of the specification.

[0049] Example 1, see Figure 1 The present invention provides a pediatric massage teaching auxiliary system based on microtremor data, and the embodiment of the present invention provides a pediatric massage teaching auxiliary system based on artificial intelligence, including:

[0050] The hand movement data during the massage process is obtained through sensors, including the hand tremor data during the massage process;

[0051] Analyze the hand tremor data and generate a feature vector including tremor amplitude features, frequency features and tremor features;

[0052] Based on the feature vector and standard massage data, determine whether it belongs to the abnormal microtremor type; analyze the feature vector based on the machine learning model and output the abnormality score;

[0053] Explanation: The machine learning model constructs a pattern recognition standard for abnormal microtremors by learning standard massage data. When the input feature vector deviates greatly from the standard, a higher abnormality score is given, and vice versa;

[0054] If the tremor is abnormal, remind the student to adjust the hand gesture, relax the muscles, and optimize the angle of force; otherwise, do not take any action (if the tremor is normal, remind the student to keep the technique stable);

[0055] It should be explained in the embodiment of the present invention that the pediatric massage teaching auxiliary system includes the following steps:

[0056] Step S11: multi-source data collection and preprocessing: collecting hand motion data during the massage process through sensors, the sensors including IMU sensors, high frame rate cameras, electromyographic sensors, muscle mechanical vibration sensors, and force sensors;

[0057] Explanation: The IMU sensor (accelerometer, gyroscope, magnetometer) is installed on the trainee's wrist, palm or finger joints to comprehensively detect the tiny displacement and rotation changes of the hand, and maintain a sampling frequency greater than or equal to 100Hz to capture high-frequency jitter information; the high-frame rate camera (at least 120FPS) is used to record the trajectory of key points of the hand, obtain visual information of micro-movements, and reduce data loss caused by occlusion through the depth camera; the electromyography sensor and muscle mechanical vibration sensor are used to detect the tiny jitter and tension of the hand muscles to help determine whether the movement is caused by fatigue or tension; the force sensor is used to capture the tiny force fluctuations of the palm during the massage process to reveal the smoothness of the technique;

[0058] Step S12: Signal processing and micro-tremor feature extraction: Based on the Kalman filter technology, the interference of large-amplitude movement in the hand motion data is removed, and the micro-tremor signal of small shaking is retained. The frequency domain analysis is performed on the micro-tremor signal after the main motion signal is removed. The fast Fourier transform or wavelet transform time-frequency analysis method is applied to perform frequency division processing on the micro-tremor signal to capture the micro-tremor component in a specific frequency band. The target frequency band is usually 5Hz-10Hz, which is suitable for high-frequency micro-vibration in massage movements, and the micro-tremor characteristics of the hand in small displacements are output; a feature vector is constructed, and the feature vector includes the energy density distribution of the micro-tremor, the deviation degree of the main motion direction, and the time series correlation parameter;

[0059] Step S13: Classification of micro-tremor patterns and smoothness scoring: Use a machine learning model (such as LSTM, deep convolutional neural network) to analyze the feature vector, output an abnormal score, establish a normal micro-tremor standard by learning standard massage data, and output an abnormal score threshold; when the abnormal score exceeds the abnormal score threshold, it is determined to be an abnormal micro-tremor type, otherwise it is a normal micro-tremor type; if it is determined to be an abnormal micro-tremor type, prompt the trainee to adjust the gesture, relax the muscles, and optimize the force angle, otherwise no measures will be taken;

[0060] Explanation: The machine learning model learns the statistical distribution and timing characteristics of microtremors during normal massage during the training phase, and generates a higher abnormality score when the input feature vector deviates from the normal distribution;

[0061] In a possible embodiment, in step S11, a cross-modal data association algorithm, such as dynamic time warping (DTW) or an adaptive time correction algorithm based on mutual information, is used to perform real-time synchronization and dynamic calibration on the collected multi-source data; global alignment is performed by collecting preset synchronization marker signals (such as periodic pulses or external timestamps), and a data-driven delay estimation method is used to reduce the timing error between sensors.

[0062] In a possible embodiment, in step S11, in order to verify and quantify the subjective perception of microtremors, the recipient's feedback (such as comfort score) and physiological signals (such as heart rate, skin electrical response) are aligned with the multidimensional data provided by the sensor.

[0063] In a possible embodiment, in step S12, a sliding window analysis is used to calculate the root mean square value, variance, and standard deviation statistical features of each window to quantify the stability of the hand in each time period, and a feature vector is constructed based on the statistical features.

[0064] Background technology: The traditional Kalman filtering method may weaken the micro-tremor signal when removing the main motion signal, resulting in the inability to effectively retain the micro-tremor component of the target frequency band (such as 5Hz-10Hz) in the subsequent analysis, affecting the accurate extraction and analysis of the micro-tremor signal; by combining multi-dimensional time-frequency analysis and improved adaptive filtering technology, it is possible to analyze the spectrum characteristics in real time and dynamically adjust the filter parameters, effectively avoiding excessive weakening of the micro-tremor signal in the Kalman filter and retaining the key micro-tremor component; the adaptive filtering technology adjusts the gain and cutoff frequency according to the real-time spectrum data and noise level, thereby ensuring that the micro-tremor signal is retained to the greatest extent while removing large-scale motion and noise;

[0065] Therefore, in a possible embodiment, in step S12, the hand motion data to be processed is obtained, low-pass / band-pass filtering and multi-dimensional time-frequency analysis are performed, improved adaptive filtering (filter signal and residual information are output) is implemented based on real-time spectrum data, multi-scale reconstruction technology is used to restore the time domain details of micro-tremor and quantify the residual, a sliding window is used to extract multi-dimensional statistical features and weighted fusion is performed, and the abnormal micro-tremor pattern is determined by comparison with standard massage data, and the filter parameters are optimized through online feedback closed-loop.

[0066] In a possible embodiment, an adaptive online learning algorithm is introduced to establish a dynamic individualized normal tremor baseline, and a clustering algorithm is used to adjust the abnormality score threshold.

[0067] Summary: The embodiment of the present invention adopts multi-sensor data acquisition and cross-modal real-time synchronization technology (including IMU sensors, high frame rate cameras, electromyography and force sensors, etc.), combined with Kalman filtering, time-frequency analysis (fast Fourier transform or wavelet transform) and sliding window statistical analysis methods, to construct a feature vector based on micro-tremor amplitude, frequency and related timing parameters, and uses machine learning models (such as LSTM, deep convolutional neural network) to classify the micro-tremor pattern and score the stability of the feature vector. At the same time, adaptive online learning and individualized calibration mechanisms are introduced to achieve real-time and accurate extraction and anomaly detection of hand micro-tremor data during massage, thereby solving the problems of multi-source data timing deviation, low signal-to-noise ratio, individual differences and data fusion contradictions in traditional pediatric massage teaching, and finally providing students with timely and detailed feedback on the stability of the technique and improvement suggestions.

[0068] Example 2, reference Figure 2 The structure block diagram of the massage teaching auxiliary system based on learning progress management is shown in FIG. The difference between the embodiment of the present invention and embodiment 1 is that it also includes:

[0069] The massage technique stability optimization module inputs the acquired feature vector into the stability evaluation model and outputs the manipulation stability index; the manipulation stability index quantifies the instability of the micro-tremor in the massage action and provides a basis for subsequent real-time feedback and massage learning progress evaluation;

[0070] The massage plan deviation optimization module is used to collect the strength, rhythm, angle, time, and trajectory data of the massage in real time, and compare the collected data with the preset personalized massage plan, identify the deviation and output the plan deviation set, and output the massage plan execution quality index;

[0071] Explanation: Collect and pre-process the key parameters and trajectory data of massage at each moment; extract the key points of the actual massage process and the preset plan (such as trajectory turning points and strength and rhythm mutation points), and calculate the matching degree of each key point in time and space; use the non-traditional geometric mean and exponential mapping method to calculate the key parameter matching degree of the key parameter data in the sampling frame; based on the key point matching degree and the key parameter matching degree, jointly calculate the massage plan execution quality index through the harmonic average method; this index can objectively reflect the deviation between the actual massage action and the personalized plan, and provide a quantitative basis for subsequent feedback and skill improvement.

[0072] The learning progress management module obtains the performance of each student in the learning progress, obtains the massage plan execution quality index and the technique stability index, performs linear normalization on the massage plan execution quality index and the technique stability index, calculates the learning progress evaluation coefficient by weighted summation with the task completion efficiency, and adjusts the massage teaching content based on the learning progress evaluation coefficient.

[0073] Adjusting the massage teaching content based on the learning progress evaluation coefficient means that when the learning progress evaluation coefficient is higher than the preset target value, it is considered that the current target massage task has a good execution effect, positive incentive information is fed back, and the next stage target massage task is automatically added;

[0074] When the learning progress assessment coefficient is lower than the preset target value, it indicates that there is a deviation in the execution of the target massage task. Combined with the analysis of specific indicators (for example, if the massage program execution quality index is low, it mainly indicates a deviation in the execution of the massage program; if the technique stability index is low, it focuses on indicating insufficient technique stability), corresponding improvement measures will be taken for learners, such as real-time action correction prompts, adjusting the training program or extending the training time, until the learning progress assessment coefficient reaches the expected requirements.

[0075] It needs to be further explained in the embodiment of the present invention that the stability evaluation model satisfies the following formula:

[0076] ;

[0077] Among them, S represents the technique stability index (the value range is 0~100, the higher the value, the more stable the technique), M represents the total sampling time, and j represents the sampling sequence number. is the tremor amplitude at time j, is the standard reference amplitude (the value is not 0), represents the time-frequency energy density, is the angle between the micro-tremor direction and the main motion direction, a and β are nonlinear adjustment factors;

[0078] It needs to be further explained in the embodiment of the present invention that the method for obtaining the complete index of the massage program execution is:

[0079] The key points of the trajectory curve of the execution process of the massage program and the trajectory curve of the personalized massage program are obtained respectively, wherein the key points refer to the turning points, the mutation points of the strength and the rhythm of the trajectory curve; the matching degree of the key points is calculated based on the time-space coordinates of the key points;

[0080] The key points (including trajectory inflection points, strength and rhythm mutation points) are extracted from the preset personalized massage program trajectory and the actual execution trajectory respectively. The preset and actual key point sequences are recorded as and , where each key point ; Represent the horizontal coordinate, vertical coordinate and time coordinate respectively; suppose there are N key points, i represents the sequence number, and the key point matching degree is calculated by the following formula :

[0081] ;

[0082] ;

[0083] Among them, di represents the Euclidean distance between the i-th key points;

[0084] Record the strength, rhythm, angle, time, and trajectory data of each massage plan, divide the execution process of the massage plan into several moments according to the number of sampling frames, and record the average value of strength, rhythm, angle, and time at each moment;

[0085] It represents the matching degree of the pth dimension, which includes strength, rhythm, angle, time, and trajectory data, that is, ;use represents the deviation of the pth dimension; suppose there are M sampling moments, j represents the sequence number, and the average value of intensity, rhythm, angle and time is recorded as the actual parameter sequence , the corresponding preset parameter sequence is recorded as , ;

[0086] The matching degree of key massage parameters is calculated by the following formula:

[0087] ;

[0088] ;

[0089] The matching degree of the combined key points and the matching degree of the key massage parameters are used to calculate the massage program execution quality index through the following formula;

[0090] ;

[0091] What needs to be further explained in the embodiments of the present invention is that real-time feedback and optimization suggestions are automatically generated based on the massage plan execution quality index and the technique stability index; through the visual interface, the trainee can view his current massage plan execution quality index and technique stability index in real time; according to the technique stability index, the trainee is prompted in real time to adjust the gestures, relax the muscles, and optimize the force application angle; the massage plan execution quality index is fed back to the trainee or teacher in real time to ensure the efficiency and efficacy of the massage process; for example, if the force is too great, it is suggested to reduce the force; if the rhythm is too fast or too slow, it is recommended to adjust the rhythm.

[0092] Explanation: The anomaly score quantifies the degree of deviation of the feature vector, while the technique stability index reflects the stability of the feature vector; the anomaly score focuses on detecting the degree of deviation of the current action from the normal state, and is an anomaly detection indicator; the technique stability index comprehensively evaluates the smoothness of the overall massage technique and directly reflects the quality of the massage technique; the anomaly score mainly depends on the probability distribution of the machine learning model, reflecting the degree to which the data deviates from the normal baseline; the technique stability index forms a comprehensive evaluation indicator by nonlinearly fusing multiple time-frequency and time-domain features (such as motion amplitude deviation, energy density, and direction deviation); the anomaly score is usually a normalized value from 0 to 1, which is used to determine whether it has entered an abnormal state; and the technique stability index is usually presented in a score of 0-100, which more intuitively reflects the stability of the technique.

[0093] Summary: The embodiment of the present invention, by introducing a manipulation stability evaluation module and a massage plan deviation optimization module, combined with a learning progress management module and an abnormal scoring system, realizes a comprehensive solution from real-time monitoring, evaluation, and feedback of massage movements to dynamic adjustment of learning content. Through the comprehensive analysis of the manipulation stability index and the massage plan execution quality index, the stability and execution quality of the massage movements can be accurately evaluated, thereby providing students with personalized feedback to ensure the efficiency and pertinence of massage teaching.

[0094] Embodiment 3 is different from Embodiment 2 in that it further comprises a personalized massage program generation module, and the personalized massage program generation module is used to obtain a preset personalized massage program;

[0095] The personalized massage program generation module is used to obtain the physiological data and historical efficacy data of children, and generate a personalized massage program through deep learning model training. The personalized massage program includes massage techniques, strength, rhythm, time, and trajectory parameters. The personalized massage program is obtained in the following ways:

[0096] Step 101, obtaining several groups of historical data, each group of historical data including physiological data of the child (such as age, weight, height, health status, etc.) and corresponding historical massage efficacy data (the historical massage efficacy data includes massage technique, strength, rhythm, time, trajectory and efficacy feedback), cleaning and standardizing the collected historical data, and outputting standardized historical data;

[0097] Step 102: Building a massage program generation model based on a deep learning model: Using standardized historical data and a deep learning model (such as a deep neural network or a long short-term memory network) to build an initial massage program generation model, and training based on standardized historical data to learn the mapping relationship between children's physiological data and massage efficacy, that is, predicting massage programs based on children's physiological data; using the mean square error of the massage program and the cross entropy loss of the massage efficacy as the loss function, training until the loss function If the requirements are met, the trained massage plan generation model is output;

[0098] Step 103, massage plan output and personalized recommendation: input the real-time collected information into the trained massage plan generation model, and output the corresponding personalized massage plan. The generated massage plan will be used as the standard for subsequent real-time monitoring and effect evaluation, and provide guidance for action adjustment and effect optimization during the massage process.

[0099] It is necessary to further explain in the embodiments of the present invention that the real-time collected information includes the child's current physiological data (such as weight, heart rate, body temperature, etc.) and real-time data during the massage process (such as real-time data of manipulation, strength, rhythm, time, and trajectory parameters); the real-time collected information is input into the trained massage program generation model, and the massage program generation model generates a corresponding personalized massage program according to the input data;

[0100] What needs to be further explained in the embodiment of the present invention is that the features of real-time data and physiological data during the massage process are extracted as input features, and the massage plan generation model gives a personalized massage plan by learning the impact of physiological data on the therapeutic effect.

[0101] It needs to be further explained in the embodiment of the present invention that the loss function satisfies:

[0102] ;

[0103] in, The weight coefficient of the regression task, represents the weight coefficient of the classification task, Indicates the massage plan or massage efficacy label (such as whether it is effective) predicted by the massage plan generation model. Indicates the real massage plan or massage efficacy label (such as whether it is effective); Refers to the mean square error of the massage plan, which is used to optimize the accuracy of the massage plan and help the model accurately predict the specific parameters of the massage (such as strength, rhythm, etc.). The prediction of massage techniques is a continuous variable. MSE can measure the difference between the predicted value and the true value and drive the model to learn in the direction of minimizing the error. The cross entropy loss part is used to optimize the evaluation of massage effect. Massage effect is usually classified by labels (such as effective / ineffective). The cross entropy loss part effectively measures the performance of the model in the massage effect classification task and ensures the effectiveness prediction of the massage plan.

[0104] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A pediatric massage teaching auxiliary system based on artificial intelligence, characterized in that: include: The hand movement data during the massage process is obtained through sensors, including hand tremor data during the massage process; Analyze the hand tremor data and generate a feature vector including tremor amplitude features, frequency features and tremor features; By learning the standard massage data, the abnormal scoring threshold is output; an adaptive online learning algorithm is introduced to establish a dynamic individualized normal microtremor baseline, and the abnormal scoring threshold is adjusted using a clustering algorithm; The feature vector is analyzed based on the machine learning model, and an abnormal score is output. The abnormal score is used to determine whether it belongs to the abnormal microtremor type; when the abnormal score exceeds the abnormal score threshold, it is determined to be an abnormal microtremor type; If the tremor is abnormal, remind the student to adjust the gesture, relax the muscles, and optimize the angle of force; otherwise, no action will be taken; The massage plan deviation optimization module is used to collect the strength, rhythm, angle, time, and trajectory data of the massage in real time, and compare the collected data with the preset personalized massage plan to output the massage plan execution quality index; Automatically generate real-time feedback and optimization suggestions based on the massage program execution quality index; Through the visual interface, students can view their current massage plan execution quality index in real time; according to the manipulation stability index, students are prompted to adjust their gestures, relax their muscles, and optimize the force angle in real time; the massage plan execution quality index is obtained as follows: Collect and preprocess the key parameters and trajectory data of massage at each moment; extract the key points of the actual massage process and the preset plan, and calculate the matching degree of each key point in time and space; calculate the matching degree of key parameters by using the non-traditional geometric mean and exponential mapping method for the key parameter data in the sampling frame; Based on the matching degree of key points and key parameters, the massage plan execution quality index is jointly calculated through the harmonic mean method.

2. According to the artificial intelligence-based pediatric massage teaching auxiliary system of claim 1, it is characterized in that: Based on Kalman filtering technology, the interference of large-amplitude movements in hand motion data is removed, the tremor signal is retained, and the frequency domain analysis is performed on the tremor signal after the main motion signal is removed. The fast Fourier transform or wavelet transform time-frequency analysis method is applied to perform frequency division processing on the tremor signal, capture the tremor components in the target frequency band, and output the micro-jitter features. The target frequency band is set to 5Hz-10Hz. A feature vector is constructed based on the micro-jitter features. The feature vector includes the energy density distribution of the micro-jitter, the deviation of the main motion direction, and the time series correlation parameters.

3. The artificial intelligence-based pediatric massage teaching auxiliary system according to claim 1 is characterized in that: The hand motion data to be processed is obtained, low-pass / band-pass filtering and multi-dimensional time-frequency analysis are performed, improved adaptive filtering is implemented based on real-time spectrum data, filtered signals and residual information are output, multi-scale reconstruction technology is used to restore the time domain details of micro-tremor and quantify the residual, sliding windows are used to extract multi-dimensional statistical features and weighted fusion is performed, and the abnormal micro-tremor pattern is determined by comparison with standard massage data, and the filter parameters are optimized by online feedback closed-loop.

4. The artificial intelligence-based pediatric massage teaching auxiliary system according to claim 1, characterized in that: It includes a massage technique stability optimization module, which inputs the acquired feature vector into a stability evaluation model, outputs a manipulation stability index, and automatically generates real-time feedback and optimization suggestions based on the manipulation stability index; through a visual interface, students can view their current manipulation stability index in real time; the stability evaluation model satisfies the following formula: ； Among them, S represents the technique stability index, M represents the total sampling time, and j represents the sampling sequence number. is the tremor amplitude at time j, is the standard reference amplitude, ≠0, represents the time-frequency energy density, is the angle between the micro-tremor direction and the main motion direction, and a and β are nonlinear adjustment factors.

5. The artificial intelligence-based pediatric massage teaching auxiliary system according to claim 1 is characterized in that: The method for obtaining the massage program execution quality index includes: The key points of the trajectory curve of the execution process of the massage program and the trajectory curve of the personalized massage program are obtained respectively, wherein the key points refer to the turning points, the mutation points of the strength and the rhythm of the trajectory curve; the matching degree of the key points is calculated based on the time-space coordinates of the key points; The key points are extracted from the preset personalized massage plan trajectory and the actual execution trajectory respectively, and the preset and actual key point sequences are recorded as and , where each key point ; Represent the horizontal axis, vertical axis and time coordinate respectively; Assume there are N key points, i represents the sequence number, and the key point matching degree is calculated by the following formula: : ； ； Among them, di represents the Euclidean distance between the i-th key points; Record the strength, rhythm, angle, time, and trajectory data of each massage plan, divide the execution process of the massage plan into several moments according to the number of sampling frames, and record the average value of strength, rhythm, angle, and time at each moment; It represents the matching degree of the pth dimension, which includes strength, rhythm, angle, time, and trajectory data, that is, ;use represents the deviation of the pth dimension; suppose there are M sampling moments, j represents the sequence number, and the average value of intensity, rhythm, angle and time is recorded as the actual parameter sequence , the corresponding preset parameter sequence is recorded as ,and ; The matching degree of key massage parameters is calculated by the following formula: ； ； The matching degree of the joint key points and the matching degree of the key massage parameters are calculated by the formula The massage program execution quality index was calculated.

6. The artificial intelligence-based pediatric massage teaching auxiliary system according to claim 4, characterized in that: include: The learning progress management module obtains the performance of each student in the learning progress, obtains the massage plan execution quality index and the technique stability index, performs linear normalization on the massage plan execution quality index and the technique stability index, and calculates the learning progress evaluation coefficient by weighted summation with the task completion efficiency. The massage teaching content is adjusted based on the learning progress evaluation coefficient. When the learning progress evaluation coefficient is higher than the preset target value, it is considered that the current target massage task is executed well, and positive incentive information is fed back and the next stage target massage task is added.

7. The artificial intelligence-based pediatric massage teaching auxiliary system according to claim 6, characterized in that: include: A personalized massage program generation module, which is used to obtain a preset personalized massage program; The personalized massage program generation module is used to obtain children's physiological data and historical efficacy data, build a massage program generation model based on a deep learning model, and train it through standardized historical data to learn the relationship between children's physiological data and massage efficacy; Children's information is collected in real time and input into the trained model to generate a personalized massage plan; the personalized massage plan includes massage techniques, strength, rhythm, time, and trajectory parameters.

Images