Intelligent auxiliary system and method for rehabilitation nursing

Abstract

The invention relates to the technical field of computer science, and discloses an intelligent auxiliary system and method for rehabilitation nursing, and the method comprises the steps: obtaining and inputting user data, and forming a personalized rehabilitation threshold; a reinforcement learning model is initialized, real-time physiological feedback of the user is collected, a preliminary training scheme is generated in combination with the personalized rehabilitation threshold, and the training scheme is optimized in combination with micro expressions and gestures of the user; performing rehabilitation progress evaluation on the trained user data, and judging whether a training scheme needs to be adjusted or not; and based on a judgment result, updating a training scheme, and adjusting a personalized rehabilitation target. The intelligent decision-making ability of the training strategy is improved, human intervention requirements are reduced, the rehabilitation experience and training effect of the user are finally improved, and the rehabilitation process is accelerated.

Description

Technical Field

[0001] The present invention relates to the field of computer science and technology, and in particular to an intelligent auxiliary system and method for rehabilitation nursing. Background Art

[0002] With the rapid development of artificial intelligence and Internet of Things technologies, the field of rehabilitation nursing has gradually evolved towards intelligence and personalization. Traditional rehabilitation training methods mainly rely on the experience of physicians and static rehabilitation plans, lacking dynamic monitoring and adjustment of patients' real-time physiological status and training effects. In recent years, intelligent rehabilitation systems based on sensor technology and machine learning algorithms have gradually emerged. For example, wearable devices are used to collect patients' physiological data (such as heart rate, blood oxygen saturation, etc.), and training suggestions are generated in combination with machine learning models. In addition, the introduction of micro-expression analysis and gesture recognition technology provides a richer dimension of user feedback for rehabilitation training. However, existing technologies are mostly limited to the analysis of a single data source, lack of deep integration of multimodal data (such as physiological signals, micro-expressions, gestures, etc.), and the adjustment of training programs often depends on preset rules, making it difficult to achieve true personalization and dynamic optimization. Summary of the invention

[0003] In view of the above-mentioned problems, the present invention is proposed.

[0004] Therefore, the technical problem solved by the present invention is: the technical problem that the training program is out of touch with actual needs, the user compliance is poor, and the rehabilitation effect is not good due to the lack of multimodal data fusion and dynamic personalized adjustment in the existing rehabilitation nursing technology.

[0005] In order to solve the above technical problems, the present invention provides the following technical solutions: an intelligent auxiliary method for rehabilitation nursing, comprising: Obtain and enter user data to form personalized rehabilitation thresholds; Forming a personalized rehabilitation threshold includes using the user's real-time physiological signals before training to establish the user's static physiological characteristics baseline USB; ≤MET< The low-intensity movements are collected during the training process, and the real-time physiological signals and micro-expressions of the users are collected. The key features of each movement are extracted, and the low-intensity exercise physiological characteristic curve DUS is established. It is compared with the static physiological characteristic baseline USB, and the relative change rate RCR is calculated. Among them, MET stands for metabolic equivalent, which is the unit for measuring exercise intensity. Indicates the lower limit of MET value, Indicates the upper limit of MET value; personalized rehabilitation threshold includes individualized training lower limit LBT and individualized training upper limit UBT; When RCR is lower than When RCR exceeds , is the upper limit of individual training UBT, where represents the lower threshold of individualized training, Represents the upper threshold of individualized training; Initialize the reinforcement learning model, collect the user's real-time physiological feedback and combine it with the personalized rehabilitation threshold to generate a preliminary training plan, and optimize the training plan by combining the user's micro-expressions and gestures; Evaluate the rehabilitation progress of user data after training to determine whether the training program needs to be adjusted; Based on the judgment results, update the training plan and adjust the personalized rehabilitation goals.

[0006] As a preferred solution of the intelligent auxiliary method for rehabilitation nursing described in the present invention, forming a personalized rehabilitation threshold includes using OpenFace to analyze user micro-expressions to extract the user's expression characteristics during low-intensity movements.

[0007] As a preferred solution of the intelligent auxiliary method for rehabilitation nursing described in the present invention, wherein: initializing the reinforcement learning model includes defining the state variables and action space of the reinforcement learning and setting the reward function; The state variable Including: the user's real-time physiological signals, RCR, facial features and personalized rehabilitation threshold; the action space Including: training intensity, training duration, rest interval and training type conversion; the reward function Including: training effect rewards, personalized rehabilitation threshold rewards, training compliance rewards, training goal achievement rewards and user expression feature rewards; The personalized recovery threshold reward includes: When When When, give rewards; The training compliance reward includes giving a reward when the user actively indicates through gestures that he is more willing to participate in the training; The facial expression features include concentration and pain index; when concentration And the pain index When the pain index When a penalty is imposed, represents the concentration threshold, and Indicates the pain index threshold.

[0008] As a preferred solution of the intelligent auxiliary method for rehabilitation nursing described in the present invention, generating a preliminary training program includes collecting n time steps , And the recording strategy model calculates the probability distribution of taking each action in each state ; For and and combining in time steps, the advantage function is calculated using the Generalized Advantage Estimation method ; Generating the preliminary training plan also includes constructing the policy optimization objective function and updating the policy parameters through the gradient descent method; Adopting a clipping mechanism to construct the policy optimization objective function ; Updating the policy parameters includes defining the final loss function , updating , mapping the policy parameters to the training plan, which is expressed by the formula:

[0009] where represents the probability distribution of calculating the optimal action in the current state according to the policy ; represents sampling a specific action from the probability distribution for training, is the policy parameter, representing the parameter vector of

[0010] As a preferred solution of the intelligent assistance method for rehabilitation care described in the present invention, where: Combining the user's micro-expression and gesture to optimize the training plan includes using , the expression feature, and the user's gesture as inputs, and using the IF-THEN rule to adjust; The IF-THEN rule includes: IF the user's micro-expression pain index > q and the concentration < d, THEN reduce the training intensity, and continuously observe the user's adaptation in the subsequent rounds of training, while automatically adjusting the training target to bias towards low-intensity loads. If rounds later, the pain index continues to be higher than a, then introduce active interaction and ask the user if they want to pause the training; IF the user's micro-expression pain index and the concentration , THEN increase the training duration each time in future training, while monitoring the change in the pain index. If the user's pain index increases to in subsequent training, then automatically withdraw the adjustment and restore the original duration; IF user gestures to indicate continued training or concentration , THEN the user shortens the rest interval in the next round of training. If the user's pain index increases during training, the regular rest interval is restored; in, Indicates the high threshold of the pain index, Indicates the low threshold of the pain index, represents the recovery threshold of the pain index, Indicates the low threshold of concentration, Indicates a medium threshold of concentration, Indicates a high threshold for concentration.

[0011] As a preferred solution of the intelligent auxiliary method for rehabilitation nursing described in the present invention, the rehabilitation progress evaluation includes extracting rehabilitation features from the trained user data and constructing rehabilitation progress indicators; fitting the rehabilitation progress curve with the rehabilitation progress indicators using the time series method. , every ten trainings are set as an analysis window. In an analysis window, when When p% of the data points are not within the personalized rehabilitation threshold, the training plan is updated, and p% indicates how many proportions of data points exceed the personalized rehabilitation threshold in the current analysis window.

[0012] As a preferred solution of the intelligent auxiliary method for rehabilitation nursing described in the present invention, wherein: based on the judgment result, updating the training plan includes using the user's real-time physiological feedback, , facial expression features constitute a decision adjustment matrix; use the decision adjustment matrix to reversely deduce the deviation of training intensity and update the training plan.

[0013] An intelligent auxiliary system for rehabilitation nursing, comprising: Data module, which obtains and enters user data to form personalized rehabilitation thresholds; The training program module initializes the reinforcement learning model, collects the user's real-time physiological feedback according to the personalized rehabilitation threshold, generates a preliminary training program, and optimizes the training program in combination with the user's micro-expressions and gestures; The evaluation module evaluates the rehabilitation progress of the user data after training and determines whether the training plan needs to be adjusted; The adjustment module updates the training plan and adjusts the personalized rehabilitation goals based on the judgment results.

[0014] A computer device comprises: a memory and a processor; the memory stores a computer program, wherein the processor implements the steps of any one of the methods of the present invention when executing the computer program.

[0015] A computer-readable storage medium stores a computer program, wherein the computer program, when executed by a processor, implements the steps of any one of the methods of the present invention.

[0016] Beneficial effects of the present invention: The intelligent auxiliary method for rehabilitation nursing provided by the present invention constructs a personalized rehabilitation threshold (LBT / UBT) by acquiring multimodal data of users, and combines the RCR calculation model of static physiological baseline (USB) and dynamic motion characteristics (DUS) to accurately identify the user's tolerance range and avoid training overload or insufficient intensity; adopts a reinforcement learning framework to integrate real-time physiological signals, micro-expressions and gesture interactions, and realizes multi-dimensional target optimization through state space modeling and reward function design, thereby improving training safety and user participation; constructs a policy gradient optimization objective function based on the PPO algorithm, dynamically adjusts training parameters using generalized advantage estimation, and realizes human-computer collaborative decision-making in combination with the IF-THEN rule base, significantly improving the real-time and robustness of program adjustment; analyzes the rehabilitation progress curve through time series, and reversely derives the deviation cause in combination with the decision matrix, dynamically calibrates the personalized goal, and forms an "evaluation-adjustment-verification" closed loop, ultimately realizing the triple dynamic adaptation of the rehabilitation program to the user's physiological state, psychological feedback and rehabilitation progress. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0018] Figure 1 An overall flow chart of an intelligent assistance method for rehabilitation nursing provided in the first embodiment of the present invention. DETAILED DESCRIPTION

[0019] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative work should fall within the scope of protection of the present invention.

[0020] Example 1, reference Figure 1 , as an embodiment of the present invention, provides an intelligent assistance method for rehabilitation nursing, comprising: S1: Obtain and enter user data to form a personalized rehabilitation threshold.

[0021] User data includes user name, user age, user gender, user health status, user current exercise ability, user rehabilitation goals, user real-time physiological signals, user real-time physiological feedback and overall exercise data of the rehabilitation training user.

[0022] Preprocessing the collected user data includes ensuring that the user's basic information (name, age, gender, etc.) is complete.

[0023] Missing data is removed, the format of the user's real-time physiological signals is converted, and the data units are unified. The motion data is represented by a numerical vector for subsequent calculations.

[0024] The Z-score method is used to detect abnormal data points and eliminate outliers caused by sensor failure or misoperation.

[0025] According to experience, , , , , +20%.

[0026] The formation of personalized rehabilitation thresholds includes using the user's real-time physiological signals before training to establish the user's static physiological characteristics baseline USB. Through five low-intensity movements of 1.5≤MET<3.0, the user's real-time physiological signals and user micro-expressions during training are collected, the key features of each movement are extracted, and the low-intensity exercise physiological characteristics curve DUS is established. It is compared with the static physiological characteristics baseline USB and the relative change rate RCR is calculated. The formula is:

[0027] OpenFace is used to analyze user micro-expressions and extract the user's facial expression features during low-intensity actions.

[0028] The key features include physiological signal features, motion features, and training load features.

[0029] The threshold of personalized rehabilitation includes the lower limit of personalized training LBT and the upper limit of personalized training UBT.

[0030] When RCR is lower than -10%, it is the lower limit of individualized training LBT, and when RCR exceeds +20%, it is the upper limit of individualized training UBT.

[0031] Furthermore, by collecting the user's basic information, real-time physiological signals and motion data, it is ensured that the training program can be adjusted according to individual characteristics, rather than adopting a unified fixed training standard. At the same time, through preprocessing (data format conversion, outlier removal, etc.), the integrity and reliability of the data are guaranteed, making subsequent calculations more stable and accurate. In addition, a static physiological characteristic baseline USB and a low-intensity exercise physiological characteristic curve DUS are established through a low-intensity exercise test (MET1.5≤MET<3.0), and the relative change rate RCR is calculated.

[0032] Furthermore, personalized rehabilitation thresholds (LBT, UBT) ensure that the user's training program is within their capabilities, avoiding injuries caused by excessive training loads or reduced rehabilitation efficiency due to low training intensity. In addition, through micro-expression analysis and cosine similarity feature matching, the system can comprehensively consider physiological signals, movement characteristics and subjective feelings, making training adjustments more intelligent. At the same time, abnormal data is eliminated and normalized.

[0033] S2: Initialize the reinforcement learning model, collect the user's real-time physiological feedback and combine it with the personalized rehabilitation threshold to generate a preliminary training plan, and optimize the training plan by combining the user's micro-expressions and gestures.

[0034] Initializing the reinforcement learning model includes defining the state variables and action space of reinforcement learning and setting the reward function.

[0035] The state variable Including: user's real-time physiological signals, RCR, facial features and personalized rehabilitation threshold. Including: training intensity, training duration, rest interval and training type conversion. Including: training effect rewards, personalized rehabilitation threshold rewards, training compliance rewards, training goal achievement rewards and user expression feature rewards.

[0036] The formula is:

[0037] in, Indicates that at time step Total rewards on. Indicates training effect reward. Represents a personalized recovery threshold reward. Indicates training compliance reward. Reward for achieving the training goal. Represents rewards based on user expression features.

[0038]

[0039]

[0040] in, Indicates a reward reflecting the training effect. Reward for achieving the training goal.

[0041] The personalized recovery threshold reward includes: When A penalty of -10 is imposed. , reward +5.

[0042] The training compliance reward includes a reward of +5 when the user actively indicates through gestures that he is more willing to participate in the training.

[0043] Based on experience , , .

[0044] The facial expression features include concentration and pain index. And the pain index When the pain index is , the penalty is -5.

[0045] Generating a preliminary training plan includes setting a time step to 15 seconds and collecting 10 time steps , and , record policy model , calculate each state Take the following actions The probability distribution of . 10 time steps , and Combination , using the generalized dominance estimation method to calculate the dominance function .

[0046] The formula is:

[0047] in, represents the advantage function, represents the contribution of weighted future rewards, Indicates immediate reward, represents the discounted future state value, Indicates the future state The status value of Represents the discount factor. Indicates in status The following value assessment.

[0048] Generating a preliminary training plan also includes constructing a policy optimization objective function and updating policy parameters via gradient descent.

[0049] Adopting the pruning mechanism to construct a strategy to optimize the objective function , the formula is:

[0050]

[0051] in, represents the policy optimization objective function, Represents the time step Perform expectation calculations, represents the probability ratio, Indicates that under the current strategy, in state Select Action probability. Indicates that under the old policy, in the state Select Action The probability of represents the advantage function. represents the clipping function, Indicates the clipping range.

[0052] Updating strategy parameters includes defining the final loss function ,renew , the formula is:

[0053]

[0054]

[0055] in, represents the final loss function, represents the policy optimization objective function, and represents the hyperparameter, represents the value function error term, is a strategy parameter, indicating The parameter vector of represents the learning rate, Represents the loss function Calculate the gradient, represents the policy entropy.

[0056] Mapping the strategy parameters to the training scheme is expressed as:

[0057] in, According to the strategy Calculate the current state The best action The probability distribution of . Indicates sampling a specific action according to the probability distribution Conduct training.

[0058] Optimizing training programs based on user micro-expressions and gestures includes: , facial features and user gestures as input, and use IF-THEN rules to Make adjustments.

[0059] The IF-THEN rules include: IF the user's micro-expression pain index is greater than 0.8 and the concentration is less than 0.3, THEN reduce the training intensity by 20%, and observe the user's adaptation in the subsequent three rounds of training. At the same time, automatically adjust the training target to a low-intensity load. If the pain index continues to be higher than 0.8 after three rounds, active interaction is introduced to ask the user whether to suspend training.

[0060] Based on experience , , 0.9.

[0061] IF the user's micro-expression pain index is <0.4 and concentration is >0.7, THEN increase the training time by 10% each time in the next two rounds of training, and monitor the changes in the pain index. If the user's pain index rises to >0.7 in subsequent training, the adjustment will be automatically withdrawn and the original duration will be restored.

[0062] IF the user gestures to continue training or the concentration level is > 0.9, THEN the user shortens the rest interval by 10% in the next round of training. If the user's pain index increases during training, the regular rest interval is restored.

[0063] Furthermore, by defining the state variables, action space, and reward function of the reinforcement learning model, the system can perceive the user's physiological signals, micro-expression feedback, and training compliance, thereby dynamically adjusting the training parameters within each training time step. Secondly, through the advantage function and strategy optimization objective function, the system can efficiently learn the optimal training strategy, and use a clipping mechanism to avoid excessive strategy updates and improve training stability. In addition, the IF-THEN rule is combined with micro-expression and gesture optimization to ensure that when the user's physiological or psychological state is abnormal, the system can make adaptive adjustments to avoid excessive training load or hindered rehabilitation progress.

[0064] Furthermore, through the introduction of the reinforcement learning framework, the system can autonomously learn the optimal training plan. Compared with the traditional fixed training plan, it can dynamically adapt to the user's status and improve the efficiency of rehabilitation training and user experience. Secondly, the personalized rehabilitation threshold reward mechanism ensures that the training load is always within an appropriate range, which not only avoids the risks of overtraining, but also prevents insufficient training intensity from affecting the progress of rehabilitation. Furthermore, by optimizing the training plan based on the user's micro-expressions and gestures, the system can perceive the user's real-time status and provide more personalized training adjustments, so that the user can adapt to the training rhythm both physically and mentally. In addition, the strategy optimizes the objective function and the gradient descent update strategy.

[0065] S3: Evaluate the rehabilitation progress of the user data after training to determine whether the training program needs to be adjusted.

[0066] The rehabilitation progress assessment includes extracting rehabilitation features from the user data after training and constructing rehabilitation progress indicators. The rehabilitation progress indicators are fitted with the rehabilitation progress curve using the time series method. The rehabilitation characteristics include physiological characteristics, motor characteristics and subjective characteristics.

[0067]

[0068] in, Rehabilitation progress indicator, represents the weight of physiological characteristics, represents the motion feature weight, Represents the subjective feature weight.

[0069] Arrange the rehabilitation progress indicators at each moment after training in chronological order to form a time series representation as follows:

[0070] in, Represents the rehabilitation progress indicator after the first training session. Represents the rehabilitation progress indicator after the second training session. represents the rehabilitation progress index after the nth training session, Indicates the total number of training times.

[0071] Based on experience .

[0072] The sliding average method is fitted to the series as Every ten trainings are set as an analysis window. In an analysis window, When 30% of the data points are not within the personalized recovery threshold, the training plan is updated.

[0073] Furthermore, through rehabilitation feature extraction, rehabilitation progress indicator construction, time series analysis and sliding average fitting, the user's state changes during training can be accurately tracked. The setting of the analysis window ensures the trend evaluation of rehabilitation progress, enabling the system to identify the stability of training effects and make adjustments at the appropriate time.

[0074] Furthermore, through time series analysis, the system can detect abnormal training effects in a timely manner and trigger intelligent adjustments when 30% of data points exceed the personalized rehabilitation threshold, thereby preventing users from entering an ineffective training phase and improving rehabilitation efficiency.

[0075] S4: Based on the judgment results, update the training plan and adjust the personalized rehabilitation goals. , the facial features are aligned and normalized in the same time window (10s) to form a state vector, and the state vectors of 10 time steps are combined to form a decision adjustment matrix. The decision adjustment matrix is ​​used to reversely deduce the deviation of the training intensity and update the training plan.

[0076] in, represents the state vector, represents the decision adjustment matrix, represents the current time step, represents how many past time steps are included in the decision adjustment matrix, represents the earliest time step.

[0077] Using the decision adjustment matrix and RCR value, the deviations of training intensity, training duration and rest interval can be reversely deduced: when RCR>+20%, it means that the current training intensity is too high, so the training intensity should be reduced by 20% and the training duration should be shortened.

[0078] When RCR<-10%, it means that the current training intensity is insufficient and the training time should be increased.

[0079] The reverse-derived adjustment suggestions are used as feedback and input into the reinforcement learning model to update the training plan.

[0080] Furthermore, by aligning and normalizing data within a fixed time window (10s), the system can effectively capture the short-term change trend of the user's status, and use historical data of multiple time steps (i.e., decision adjustment matrix) for analysis, thereby reversely inferring the deviation of training intensity, training duration, and rest interval. Based on this, the training plan is adjusted to keep the training process accurate, adaptive, and progressive, avoiding fatigue caused by too high training intensity or slow recovery progress caused by too low training intensity.

[0081] Furthermore, by constructing a decision adjustment matrix, the system can more accurately identify the user's physiological state and rehabilitation progress, reducing misjudgments caused by data noise or single time step judgments. Secondly, based on the threshold judgment of RCR, the training plan can be adjusted in time when the user's training load is too high or too low, reducing the risk of sports injuries while ensuring the steady progress of rehabilitation. In addition, the introduction of the reinforcement learning model enables the training plan to be continuously optimized over time, improves the intelligent decision-making ability of the training strategy, reduces the need for human intervention, and ultimately improves the user's rehabilitation experience and training effect, accelerating the rehabilitation process.

[0082] Embodiment 2 is an embodiment of the present invention, which provides an intelligent assistance system for rehabilitation nursing, including: The data module obtains and enters user data to form a personalized rehabilitation threshold.

[0083] The training program module initializes the reinforcement learning model, collects the user's real-time physiological feedback according to the personalized rehabilitation threshold, generates a preliminary training program, and optimizes the training program in combination with the user's micro-expressions and gestures.

[0084] The evaluation module evaluates the rehabilitation progress of user data after training and determines whether the training plan needs to be adjusted.

[0085] The adjustment module updates the training plan and adjusts the personalized rehabilitation goals based on the judgment results.

[0086] Embodiment 3, an embodiment of the present invention, is different from the first two embodiments in that: If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc., which can store program codes.

[0087] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in conjunction with such instruction execution systems, devices or apparatuses. For the purposes of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in conjunction with such instruction execution systems, devices or apparatuses.

[0088] More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk case (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be a paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering or, if necessary, processing in another suitable manner, and then stored in a computer memory.

[0089] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0090] Example 4 is an embodiment of the present invention, which provides an intelligent auxiliary system and method for rehabilitation nursing. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through simulation experiments.

[0091] Twenty rehabilitation subjects after hip replacement surgery were selected, with an age distribution of 45 to 70 years old, including 11 males and 9 females. These rehabilitation subjects were arranged to enter the same rehabilitation institution within seven to ten days after surgery, and received two training sessions a day, each lasting about 30 minutes, in order to achieve dynamic regulation and evaluation of their rehabilitation process through the introduction of personalized thresholds and reinforcement learning models.

[0092] Before the start of the experiment, all subjects were collected basic information, including name, age, gender, medical history, current exercise ability and rehabilitation goals, and baseline detection of real-time physiological signals was performed. The above baseline detection includes the root mean square value and frequency domain parameters of electromyographic signals (EMG), heart rate variability (HRV) and blood oxygen saturation (SpO2), and the recording time lasts for more than three minutes to extract the static physiological characteristics baseline USB. Subsequently, all subjects were required to complete five low-intensity movements of 1.5≤MET<3.0, including slow leg raising, seated stepping, simple squatting, slow walking and ankle flexibility training. During this process, facial expressions were captured using OpenFace software to extract micro-expression features such as concentration and pain index. At the same time, the low-intensity exercise physiological characteristic curve DUS was constructed by recording physiological data in real time, and compared with USB to calculate the relative change rate RCR. If the RCR is below -10%, the subject's individualized lower training limit LBT is set to a lower level to prevent excessive fatigue or injury; if the RCR exceeds +20%, the individualized upper training limit UBT is set to a more cautious threshold to avoid excessive physiological burden.

[0093] Based on the above data, the reinforcement learning model is initialized. The state variables of the model include the user's real-time physiological signals (such as EMG, HRV, etc.) and expression characteristics (concentration, pain index), as well as personalized rehabilitation thresholds (LBT, UBT) and the current RCR value; the action space covers training intensity, training duration, rest interval, and training type conversion. For example, once the system detects that the pain index is higher than 0.8 and the concentration is significantly reduced, the action options of "reducing intensity" or "shortening duration" will be triggered in the reinforcement learning model. The reward function includes five parts: training effect reward, personalized rehabilitation threshold reward, training compliance reward, training goal achievement reward, and user expression feature reward, which correspond to the degree of improvement of the user's ProgressIndex after training, whether it exceeds the threshold range, the patient's active willingness, the completion of the staged goal, and the state reflected by the micro-expression. After that, the training time step is set to fifteen seconds, each training round lasts for ten time steps, and a total of one hundred and fifty seconds to complete a reinforcement learning strategy update. After completing several rounds of training, the system calculates the advantage function through the generalized advantage estimation method, and then uses the gradient descent method to update the strategy parameters, combined with the clipping mechanism to stabilize the optimization process of reinforcement learning.

[0094] During the implementation process, in order to make micro-expressions and gestures better play a role in training optimization, several IF-THEN rules were set: when the pain index exceeds 0.8 and the concentration is lower than 0.3, the system automatically reduces the training intensity by 20% and observes the next three rounds of training; when the pain index is lower than 0.4 and the concentration is higher than 0.7, the training time will be increased by 10% each time in the next two rounds, but if the pain index rises to greater than 0.7, the adjustment will be withdrawn immediately. If the user's gesture clearly indicates to continue training or the concentration is detected to be higher than 0.9, the rest interval will be shortened by 10% in the next round of training. When evaluating the rehabilitation progress of user data after training, the time series method is used to fit the rehabilitation progress curve, and it is checked whether 30% of the data points exceed the personalized rehabilitation threshold range in every ten training sessions to determine whether the training plan needs to be updated again or the LBT and UBT need to be adjusted.

[0095] After the test, the main data were recorded as follows: The initial ProgressIndex values ​​of the experimental subjects ranged from 41.20 to 42.90, with an average value of 42.15 and a standard deviation of about 1.08. After one month of application of the method of the present invention, the average ProgressIndex detected rose to 71.56, with a minimum value of 69.20, a maximum value of 73.45, and a standard deviation of about 1.85. The initial ProgressIndex value of the traditional method (fixed training program) group for reference was 42.05 on average, and the final average value detected after the same length of time was 62.34, with a minimum value of 59.90, a maximum value of 64.10, and a standard deviation of about 1.66. The pain index was averaged under the reinforcement learning optimization. The average decrease was 0.33, from the initial 1.05 to 0.72, with the minimum decrease of 0.28 and the maximum decrease of 0.38; the average decrease in the control group was only 0.15; the number of times the user actively gestured to express his willingness to train reached an average of 3.20 times in every ten training sessions in the method group of the present invention, while it was only 1.54 times in the control group; during the training process, it was observed that the time period with concentration above 0.80 increased by 17.50% cumulatively, and the time period with pain index exceeding 0.80 decreased by 22.40% cumulatively; in the last three training sessions, the system detected that the RCR of two subjects was greater than +20% for a long time, so it automatically reduced the training intensity by 15% to 20% and extended the rest interval by 10.00%.

[0096] It can be seen from the above data that the personalized rehabilitation threshold and reinforcement learning model used in this embodiment realize flexible adjustment of the rehabilitation training process, and show significant advantages when compared with the traditional fixed scheme. First, the average Progress Index increased from 42.15 to 71.56 in the experimental subjects, an increase of more than 29 percentage points, while the control group only increased from 42.05 to 62.34, an increase of about 20 percentage points. In terms of magnitude, the progress brought about by the method of the present invention is obviously more obvious, indicating that personalized rehabilitation thresholds and adaptive training strategies can more effectively tap the user's physical potential and avoid ineffective or excessive training.

[0097] Further analysis shows that the significant decrease in the pain index is one of the important advantages of the method of the present invention compared to traditional methods. The average decrease in the method group was 0.33, while the control group was only 0.15. Combined with micro-expression analysis, it can be inferred that the reinforcement learning model immediately reduces the training intensity after perceiving that the user's pain index is higher than 0.8 and the concentration is lower than 0.3, which greatly reduces the user's training discomfort. In traditional fixed training programs, fine-tuning is usually performed during routine inspections, and the amplitude may be too large or the timing is not timely, resulting in low user acceptance of the training program, which in turn affects the overall rehabilitation effect.

[0098] At the same time, the number of times the user actively gestured to indicate that they were more willing to train was as high as 3.20 times per ten training sessions in the group using the method of the present invention, which is almost doubled compared to the 1.54 times in the control group. This shows that when the system can refine or call back the training plan in a timely manner according to the user's fatigue, pain index and concentration, the user's training compliance is significantly improved, and the continuity and durability of the training are enhanced. Especially with the cooperation of the IF-THEN rule, if it is detected that the pain index continues to be high, the system will also actively ask the user whether it needs to interrupt or change the action. This level of meticulous human-computer interaction allows users to gain more security and initiative during the training process.

[0099] Furthermore, both high values ​​(over +20%) and low values ​​(below -10%) in the RCR data are controlled in a timely manner. Under the method of the present invention, if the RCR is greater than +20% for a long time, the system will immediately reduce the training intensity within the range (15% to 20%) and extend the rest interval. The control group lacks timely monitoring and response, which may cause some trainees to passively persist due to fatigue and joint pain, resulting in a greater increase in the pain index. It can be seen that the method of the present invention helps to balance the training effect and patient comfort and reduce unnecessary risks. Increasing the training time in a state of high concentration can also maximize the training benefits when the user is in good physical and mental state.

[0100] In terms of algorithm operation and application scenarios, through multiple rounds of iterative training of the reinforcement learning model (PPO algorithm), it is shown that the user's fluctuating data can be effectively captured at different times, and the training effect reward, rehabilitation threshold reward and expression feature reward are taken into account in the strategy optimization objective function. It can be seen that the algorithm framework has sufficient adaptive ability to cope with individual differences. Compared with traditional methods, the intelligent auxiliary method in this embodiment proves its creativity and novelty: it can not only adjust the rehabilitation plan in real time and dynamically, but also fully considers human-computer interaction factors such as micro-expressions and gestures, making the training process more humane and effectively avoiding the shortcomings of fixed plans. Data analysis shows that the recovery efficiency and comfort of users' physical functions have been improved, showing the distinct advantages of personalization, intelligence and sustainability, and can provide reliable technical support for subsequent large-scale promotion and application.

[0101] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. An intelligent auxiliary method for rehabilitation nursing, characterized in that: include: Obtain and enter user data to form personalized rehabilitation thresholds; Forming a personalized rehabilitation threshold includes establishing a user's static physiological characteristic baseline USB using the user's real-time physiological signals before training; Through x ≤MET< The low-intensity movements are collected during the training process, and the real-time physiological signals and micro-expressions of the users are collected. The key features of each movement are extracted, and the low-intensity exercise physiological characteristic curve DUS is established. It is compared with the static physiological characteristic baseline USB, and the relative change rate RCR is calculated. Among them, MET stands for metabolic equivalent, which is the unit for measuring exercise intensity. Indicates the lower limit of MET value, Indicates the upper limit of MET value; personalized rehabilitation threshold includes individualized training lower limit LBT and individualized training upper limit UBT; When RCR is lower than When RCR exceeds , is the upper limit of individual training UBT, where represents the lower threshold of individualized training, Represents the upper threshold of individualized training; Initialize the reinforcement learning model, collect the user's real-time physiological feedback and combine it with the personalized rehabilitation threshold to generate a preliminary training plan, and optimize the training plan by combining the user's micro-expressions and gestures; Evaluate the rehabilitation progress of user data after training to determine whether the training program needs to be adjusted; Based on the judgment results, update the training plan and adjust the personalized rehabilitation goals.

2. The intelligent assistance method for rehabilitation nursing according to claim 1, characterized in that: Forming a personalized rehabilitation threshold includes using OpenFace to analyze the user's micro-expressions and extract the user's facial expression characteristics during low-intensity movements.

3. The intelligent assistance method for rehabilitation nursing according to claim 2, characterized in that: Initializing the reinforcement learning model includes defining the state variables and action space of reinforcement learning and setting the reward function; The state variable Includes: user's real-time physiological signals, RCR, facial features and personalized rehabilitation threshold; action space Including: training intensity, training duration, rest interval and training type conversion; the reward function Including: training effect rewards, personalized rehabilitation threshold rewards, training compliance rewards, training goal achievement rewards and user expression feature rewards; The personalized recovery threshold reward includes: When When When, give rewards; The training compliance reward includes giving a reward when the user actively indicates through gestures that he is more willing to participate in the training; The facial expression features include concentration and pain index; when concentration And the pain index When the pain index When a penalty is imposed, represents the concentration threshold, and Indicates the pain index threshold.

4. The intelligent auxiliary method for rehabilitation nursing according to claim 3, characterized in that: Generating a preliminary training plan involves collecting n time steps , and , record policy model , calculate each state Take the following actions The probability distribution of ; n time steps , and Combination , using the generalized dominance estimation method to calculate the dominance function ; Generating a preliminary training plan also includes constructing a policy optimization objective function and updating policy parameters by gradient descent; Adopting the pruning mechanism to construct a strategy to optimize the objective function ; Updating strategy parameters includes defining the final loss function ,renew , mapping the strategy parameters to the training scheme, the formula is expressed as: ； in, According to the strategy Calculate the current state The best action The probability distribution of Indicates sampling a specific action according to the probability distribution Conduct training, is a strategy parameter, indicating The parameter vector of .

5. The intelligent auxiliary method for rehabilitation nursing according to claim 4, characterized in that: Optimizing training programs based on user micro-expressions and gestures includes: , facial features and user gestures as input, and use IF-THEN rules to Make adjustments; The IF-THEN rule includes: IF the user's micro-expression pain index > q and the concentration < d, THEN reduce the training intensity, and continuously observe the user's adaptation in subsequent rounds of training. At the same time, automatically adjust the training goal to be biased towards low-intensity loads. If the pain index continues to be higher than a after IF users’ micro-expression pain index And concentration , THEN will increase the training time each time in the future training, and monitor the changes in the pain index at the same time. If the user's pain index increases to , the adjustment will be automatically withdrawn and the original duration will be restored; IF user gestures to indicate continued training or concentration , THEN the user shortens the rest interval in the next round of training. If the user's pain index increases during training, the regular rest interval is restored; in, Indicates the high threshold of the pain index, Indicates the low threshold of the pain index, represents the recovery threshold of the pain index, Indicates the low threshold of concentration, Indicates a medium threshold of concentration, Indicates a high threshold for concentration.

6. The intelligent assistance method for rehabilitation nursing according to claim 5, characterized in that: The rehabilitation progress assessment includes extracting rehabilitation features from the user data after training and constructing rehabilitation progress indicators; fitting the rehabilitation progress curve with the rehabilitation progress indicators using the time series method. , every ten trainings are set as an analysis window. In an analysis window, when When p% of the data points are not within the personalized rehabilitation threshold, the training plan is updated, and p% indicates how many proportions of data points exceed the personalized rehabilitation threshold in the current analysis window.

7. The intelligent assistance method for rehabilitation nursing according to claim 6, characterized in that: Based on the judgment results, the training plan is updated, including using the user's real-time physiological feedback, , facial expression features constitute a decision adjustment matrix; use the decision adjustment matrix to reversely deduce the deviation of training intensity and update the training plan.

8. An intelligent assistance system for rehabilitation nursing using the method according to any one of claims 1 to 7, characterized in that: Data module, which obtains and enters user data to form personalized rehabilitation thresholds; The training program module initializes the reinforcement learning model, collects the user's real-time physiological feedback according to the personalized rehabilitation threshold, generates a preliminary training program, and optimizes the training program in combination with the user's micro-expressions and gestures; The evaluation module evaluates the rehabilitation progress of the user data after training and determines whether the training plan needs to be adjusted; The adjustment module updates the training plan and adjusts the personalized rehabilitation goals based on the judgment results.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the intelligent assistance method for rehabilitation nursing described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the intelligent assistance method for rehabilitation nursing described in any one of claims 1 to 7 are implemented.

Images