Intelligent rehabilitation training system based on multi-parameter detection and control method

Abstract

The invention discloses an intelligent rehabilitation training system based on multi-parameter detection and a control method, and relates to the technical field of intelligent rehabilitation training, the system comprises a user information acquisition unit, a multi-source parameter acquisition unit, a rehabilitation training plan customization unit, a feedback and control unit and a quantitative evaluation unit; multi-source detection parameters corresponding to movement, physiology and environment of a target user are collected in real time through a preset sensor, the health-related data and the multi-source detection parameters are fused, a rehabilitation training domain knowledge graph is accessed to construct a training scheme recommendation model, and a personalized rehabilitation training plan of the target user is output; according to the achievable degree of the target user for each rehabilitation training action, real-time feedback is provided through a visual interface and voice prompt, and the target user is guided to adjust the training action; a brand new solution is provided for rehabilitation training based on multi-parameter detection, the training state of the user can be monitored in real time, a personalized training scheme is generated, and the rehabilitation effect and efficiency are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent rehabilitation training, and more specifically, to an intelligent rehabilitation training system and control method based on multi-parameter detection. Background Art

[0002] With the increase of population aging and chronic diseases, the importance of rehabilitation training in the field of medical health has become increasingly prominent. Traditional rehabilitation training methods mainly rely on the manual guidance of physical therapists and the self-awareness of patients, and there are problems such as difficult quantification of training effects, insufficient personalization, and uneven resource allocation. The traditional rehabilitation training methods have the following main problems: relying on manual intervention: the training process requires the full guidance of physical therapists, with high labor costs and difficulty in large-scale implementation; lack of quantitative evaluation: the training effects mainly rely on subjective evaluation and lack objective data support; insufficient personalization: the training programs are usually general-purpose and difficult to adjust according to the individual differences of patients; lack of real-time feedback: it is impossible to monitor the training status of patients in real time, and it is difficult to avoid incorrect movements or overtraining; uneven resource allocation: high-quality rehabilitation resources are concentrated in cities and large hospitals, and it is difficult for patients in remote areas to obtain timely treatment.

[0003] In recent years, with the rapid development of sensor technology, artificial intelligence (AI) and the Internet of Things (IoT), intelligent rehabilitation training systems based on multi-parameter detection have gradually become a research hotspot. Multi-parameter detection technology can collect the physiological and movement data of patients in real time by integrating multiple sensors (such as inertial sensors, electromyography sensors, heart rate sensors, etc.), providing a data basis for intelligent rehabilitation training. These data are uploaded to the cloud or local server through wireless transmission technology (such as Bluetooth, Wi-Fi), providing support for subsequent data analysis and intelligent decision-making. Therefore, how to monitor the training status of patients in real time based on multi-parameter detection and generate customized training plans according to the individual differences of patients is an urgent problem to be solved. Summary of the Invention

[0004] In order to solve the above technical problems, the present invention proposes an intelligent rehabilitation training system and control method based on multi-parameter detection, which can monitor the training status of patients in real time and generate personalized training programs, significantly improving the rehabilitation effect and efficiency.

[0005] In the first aspect of the present invention, an intelligent rehabilitation training system based on multi-parameter detection is provided, including a user information acquisition unit, a multi-source parameter acquisition unit, a rehabilitation training plan customization unit, a feedback and control unit, and a quantitative evaluation unit; The user information acquisition unit is responsible for obtaining the health-related data of the target user, preprocessing the health-related data, and constructing a health profile; The multi-source parameter acquisition unit is responsible for collecting multi-source detection parameters corresponding to the target user's motion, physiology, and environment in real time through preset sensors, and performing data preprocessing on the multi-source detection parameters; The rehabilitation training plan customization unit is responsible for integrating the health portrait of the target user and the multi-source detection parameters, obtaining the enhanced health portrait of the target user, accessing the knowledge graph in the field of rehabilitation training to construct a training plan recommendation model, using the enhanced health portrait as the model input, and obtaining the personalized rehabilitation training plan for the target user; The feedback and control unit is responsible for obtaining the achievable degree of each rehabilitation training action of the target user based on the personalized rehabilitation training plan of the target user, providing real-time feedback through a visual interface and voice prompts according to the achievable degree, and guiding the target user to adjust the training action; The quantitative evaluation unit is responsible for quantitatively evaluating the adaptability of the personalized rehabilitation training plan according to the multi-source detection data and health portrait of the target user during training, and adjusting the personalized rehabilitation training plan according to the adaptability.

[0006] In this solution, in the user information acquisition unit, health-related data of the target user is obtained, data preprocessing is performed on the health-related data, and a health portrait is constructed. Specifically: Medical text data is obtained as health-related data according to the identity information of the target user, the medical text data is standardized, and the standardized medical text data is respectively vector-encoded at different granularities using the RoBERTa pre-trained model and the Word2vec pre-trained model; The character vectors and word vectors corresponding to the medical text data are obtained, the character vectors and word vectors are concatenated in sequence to generate a fused vector representation, a Bi-GRU network is used to perform bidirectional feature extraction on the fused vector representation, and the obtained forward feature information and backward feature information are fused to output a feature vector; Using the Bi-GRU network as the encoder module of the noise reduction encoder structure, the feature vector is perturbed by adding random noise, the hidden feature distribution is obtained through linear transformation and activation function, and the decoder module is used to reconstruct the hidden feature distribution to obtain a reconstructed feature vector; The reconstructed feature vector and the feature vector are fused, and the attention mechanism is used to perform position-weighted representation on the fused feature vector; The label data representing health is obtained through big data methods, the label classification layer is trained using the label data, the position-weighted fused feature vector is classified and predicted, the label that meets the probability threshold is obtained, and the health portrait of the target user is constructed based on the obtained label.

[0007] In this solution, in the multi-source parameter acquisition unit, multi-source detection parameters corresponding to the movement, physiology, and environment of the target user are collected in real time through preset sensors. Specifically: Retrieve rehabilitation training instances according to the health profile of the target user, decompose the training actions of the rehabilitation training instances, count the occurrence frequencies of each training action in different instances, and select a preset number of training actions as basic actions based on the occurrence frequencies; Conduct an initial detection of the target user through the basic actions, obtain the movement parameters, physiological parameters, and environmental parameters of the target user during the completion of the basic actions, align the data according to the time stamp and perform data cleaning to obtain the preprocessed multi-source detection parameters.

[0008] In this solution, in the rehabilitation training plan customization unit, connect to the knowledge graph in the rehabilitation training field to construct a training plan recommendation model. Specifically: Extract the interaction information between the user's health profile and rehabilitation training actions from the rehabilitation training instances to construct an interaction graph of the user's health profile - rehabilitation training actions, obtain the knowledge graph in the rehabilitation training field, combine the interaction graph of the user's health profile - rehabilitation training actions with the knowledge graph in the rehabilitation training field to construct a training plan recommendation model, and generate a collaborative knowledge graph for rehabilitation training; Use the rehabilitation training actions that interact with the enhanced health profile of the target user to spread in the collaborative knowledge graph, find the tail entity according to the triples in the collaborative knowledge graph, and obtain a multi-layer relevant entity set of the enhanced health profile of the target user through iterative propagation; Construct a triple set of the enhanced health profile of the target user at each layer from the relevant entity set obtained at each layer, calculate the relevance score of all rehabilitation training actions and each triple, convert the relevance score into a probability, use the probability to perform weighted summation on the tail entity in the triple, and splice the results obtained after multiple propagations to obtain the initial representation of the enhanced health profile of the target user; Obtain a collaborative user health profile similar to the enhanced health profile of the target user in the collaborative knowledge graph of rehabilitation training, use a self-attention network to enhance the collaborative user health profile, and fuse to obtain a collaborative user representation; Fuse the initial representation of the enhanced health profile of the target user with the collaborative user representation to obtain an enhanced representation of the enhanced health profile of the target user, and perform convolutional aggregation using the collaborative knowledge graph of rehabilitation training to obtain a high-order representation of the rehabilitation training actions. Calculate the inner product according to the enhanced representation of the enhanced health profile of the target user and the high-order representation of the rehabilitation training actions to obtain the recommended rehabilitation training actions.

[0009] In this solution, in the rehabilitation training plan customization unit, obtain the personalized rehabilitation training plan of the target user. Specifically: Obtain the recommended rehabilitation training actions for the target user output by the constructed training plan recommendation model, encode the recommended rehabilitation training actions, randomly generate an initial combination of rehabilitation training actions, and use an improved genetic algorithm to optimize the rehabilitation training plan; Construct a fitness function based on the average rehabilitation effect corresponding to the rehabilitation training actions, calculate the fitness corresponding to different combinations of rehabilitation training actions, and select a preset number of combinations of rehabilitation training actions for replication according to the fitness to determine the search direction; During the iteration process, select the combination of rehabilitation training actions with the highest fitness for crossover recombination, and change the encoding of the rehabilitation training actions during the recombination process. When the iteration times are met, output the optimal combination of rehabilitation training actions, and construct a personalized rehabilitation training plan for the target user based on the optimal combination of rehabilitation training actions combined with the physiological parameters of the target user.

[0010] In this solution, in the feedback and control unit, obtain the achievability of the target user for each rehabilitation training action based on the personalized rehabilitation training plan of the target user. Specifically: Collect multi-source detection parameters of the target user based on the personalized rehabilitation training plan, and read the standard action parameter matrix of the rehabilitation training actions in the personalized rehabilitation training; Extract action features through the motion parameter sequence in the multi-source detection parameters, identify the rehabilitation training actions at different timestamps, obtain the motion parameter subsequences with rehabilitation training action labels, and construct the motion parameter matrix of the motion parameter subsequences; Calculate the residual matrix of the motion parameter matrix and the standard parameter matrix, obtain the achievability of the target user for each rehabilitation training action based on the residual matrix. When the achievability is less than the preset threshold, provide real-time feedback through the visualization interface and voice prompt to guide the target user to adjust the training actions; In this solution, in the quantitative evaluation unit, quantitatively evaluate the adaptability of the personalized rehabilitation training plan according to the multi-source detection data and health portrait of the target user during training. Specifically: Retrieve rehabilitation training instances according to the health portrait of the target user, set physiological parameter thresholds under different training durations based on the rehabilitation training instances, adjust the physiological parameter thresholds through the environmental parameters in the multi-source detection parameters, and use the adjusted physiological parameter thresholds to mark the physiological parameter abnormal points of the target user during the training process; Obtain and mark the rehabilitation training actions with achievability less than the preset threshold during the training process of the target user, segment the training duration, and respectively obtain the density of physiological parameter abnormal points and the marked rehabilitation training actions in different training duration segments; Characterize the fitness score of the target user and the personalized rehabilitation training plan according to the density, preset different score threshold intervals for different training duration segments. When the fitness score does not meet the preset score threshold interval, it is regarded as a mismatched segment. When the number of mismatched segments reaches the preset quantity threshold, adjust the personalized rehabilitation training plan.

[0011] The second aspect of the present invention provides an intelligent rehabilitation training control method based on multi-parameter detection, which is applied to an intelligent rehabilitation training system based on multi-parameter detection, and includes the following steps: Obtain the health-related data of the target user, and collect multi-source detection parameters corresponding to the movement, physiology and environment of the target user in real time through preset sensors, and perform data preprocessing on the health-related data and multi-source detection parameters. Fuse the health portrait of the target user and the multi-source detection parameters, obtain the fusion features, access the knowledge graph in the field of rehabilitation training to construct a training plan recommendation model, and use the fusion features as the model input to obtain the personalized rehabilitation training plan of the target user. Based on the personalized rehabilitation training plan of the target user, obtain the achievable degree of the target user for each rehabilitation training action, and provide real-time feedback through a visual interface and voice prompts according to the achievable degree to guide the target user to adjust the training action. Quantitatively evaluate the fitness of the personalized rehabilitation training plan according to the multi-source detection data and health portrait of the target user during training, and adjust the personalized rehabilitation training plan according to the fitness.

[0012] Compared with the prior art, the beneficial effects of the present invention are as follows: According to the individual differences of patients, the present invention uses intelligent algorithms to analyze health-related data and multi-source detection data to generate customized training plans, monitors the training status of users in real time through multi-parameter detection, and corrects incorrect actions in a timely manner; through data analysis, objectively evaluates the training effect, and long-term tracks the rehabilitation progress of patients. In addition, patients can train at home, reducing the number of trips to the hospital and lowering medical costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] In order to more clearly illustrate the technical solutions in the embodiments or exemplary examples of the present invention, the following will briefly introduce the drawings required for use in the embodiments or exemplary descriptions. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained according to the drawings shown without creative efforts.

[0014] Figure 1 Shows a block diagram of an intelligent rehabilitation training control system based on multi-parameter detection.

[0015] Figure 2 The flowchart of constructing a health profile in the user information collection unit is shown; Figure 3 The flowchart of constructing a training plan recommendation model in the rehabilitation training plan customization unit is shown; Figure 4 The flowchart of an intelligent rehabilitation training control method based on multi-parameter detection is shown. Detailed implementation manners

[0016] In order to be able to more clearly understand the above objects, features and advantages of the present invention, the present invention will be further described in detail below in conjunction with the accompanying drawings and specific implementation manners. It should be noted that, without conflict, the embodiments of the present application and the features in the embodiments can be combined with each other.

[0017] Many specific details are set forth in the following description in order to fully understand the present invention. However, the present invention may also be implemented in other ways different from those described herein. Therefore, the protection scope of the present invention is not limited by the specific embodiments disclosed below.

[0018] Figure 1 The block diagram of an intelligent rehabilitation training control system based on multi-parameter detection is shown.

[0019] As Figure 1 shown, in the first embodiment of the present invention, an intelligent rehabilitation training control system based on multi-parameter detection is provided, including: a user information collection unit 101, a multi-source parameter collection unit 102, a rehabilitation training plan customization unit 103, a feedback and control unit 104, and a quantitative evaluation unit 105; The user information collection unit 101 is responsible for obtaining health-related data of a target user, preprocessing the health-related data, and constructing a health profile; The multi-source parameter collection unit 102 is responsible for collecting multi-source detection parameters corresponding to the movement, physiology, and environment of the target user in real time through preset sensors, and preprocessing the multi-source detection parameters; The rehabilitation training plan customization unit 103 is responsible for fusing the health profile of the target user and the multi-source detection parameters, obtaining an enhanced health profile of the target user, accessing a knowledge graph in the field of rehabilitation training to construct a training plan recommendation model, using the enhanced health profile as the model input, and obtaining a personalized rehabilitation training plan for the target user; The feedback and control unit 104 is responsible for obtaining the achievable degree of each rehabilitation training action of the target user based on the personalized rehabilitation training plan of the target user, providing real-time feedback through a visual interface and voice prompts according to the achievable degree, and guiding the target user to adjust the training action; The quantization evaluation unit 105 is responsible for quantifying and evaluating the fitness of the personalized rehabilitation training plan according to the multi-source detection data and health profile of the target user during training, and adjusting the personalized rehabilitation training plan according to the fitness.

[0020] It should be noted that in the intelligent rehabilitation training control system, physiological, motion, and environmental data of patients are collected through sensors. Common sensor types include inertial sensors, electromyography sensors, heart rate sensors, blood oxygen sensors, and environmental sensors, etc. The collected data types include motion parameters such as joint angles, motion speeds, and accelerations, physiological parameters such as muscle activity signals, heart rate, blood pressure, and blood oxygen, and environmental parameters such as temperature, humidity, and light. The collected data is cleaned, filtered, and feature-extracted, and the data is transmitted to subsequent units for data processing through wireless technologies such as Bluetooth and Wi-Fi. After analyzing the data, a personalized training plan is generated, real-time feedback is provided through a visual interface, voice prompt, or mechanical device, and data storage, remote monitoring, and doctor-patient interaction are carried out through the cloud platform and the mobile terminal.

[0021] Figure 2 The flowchart of constructing a health profile in the user information collection unit is shown.

[0022] According to an embodiment of the present invention, in the user information collection unit, health-related data of the target user is obtained, and the health-related data is preprocessed to construct a health profile. Specifically: S202, obtain medical text data as health-related data according to the identity information of the target user, perform standardization processing on the medical text data, and perform vector encoding of different granularities on the standardized medical text data using the RoBERTa pre-trained model and the Word2vec pre-trained model respectively; S204, obtain the character vector and word vector corresponding to the medical text data, splice the character vector and word vector in sequence to generate a fused vector representation, use a Bi-GRU network to perform bidirectional feature extraction on the fused vector representation, and fuse the obtained forward feature information and backward feature information to output a feature vector; S206, use a Bi-GRU network as the encoder module of the noise reduction encoder structure, perturb the feature vector by adding random noise, obtain the hidden feature distribution through linear transformation and activation function, and use the decoder module to reconstruct the hidden feature distribution to obtain a reconstructed feature vector; S208, fuse the reconstructed feature vector and the feature vector, and use the attention mechanism to perform position-weighted representation on the fused feature vector; S210. Obtain label data representing health through big data methods, use the label data to train a label classification layer, classify and predict the fused feature vectors with position weighting, obtain labels that meet the probability threshold, and construct a health profile of the target user based on the obtained labels.

[0023] It should be noted that historical medical record information is obtained according to the identity information of the target user, medical text data is extracted from the historical medical record data as health-related data, and operations such as word segmentation and stop word removal are performed. Word segmentation is performed separately in units of characters and words, and different granularity vector encodings are performed using the RoBERTa pre-trained model and the Word2vec pre-trained model to improve the accuracy and comprehensiveness of the health label classification of the target user. The vectors of different granularities obtained by encoding are concatenated to improve the feature expression ability through multi-granularity information fusion. The Bi-GRU network is used to perform bidirectional feature extraction on the fused vector representation, and a Dropout layer is added to the Bi-GRU network structure to prevent overfitting. It is used as the encoder module of the denoising encoder network to extract the fused vector features and reduce noise interference, ensuring the coherence and robustness of semantic information. The reconstructed feature vector is obtained by reconstruction in the decoder module of the denoising encoder network, and the parameters are continuously optimized through the backpropagation algorithm to minimize the reconstruction error. After the decoder, an attention mechanism is introduced to obtain the weights of different positions of the features for feature enhancement and improve the performance of the model. Label data representing health is obtained through big data methods, such as basic attributes like age and gender, disease attributes like disease type and disease severity, and exercise attributes like exercise ability. In the label classification layer, linear transformation is performed using a fully connected layer, and then non-linear transformation is performed through the activation function ReLU to learn the feature representation, convert the output vector into the probability of the health label category, and aggregate the obtained labels to construct the health profile of the target user.

[0024] It should be noted that in the multi-source parameter acquisition unit, multi-source detection parameters corresponding to the movement, physiology, and environment of the target user are collected in real time through preset sensors. Retrieve rehabilitation training examples according to the health profile of the target user, decompose the training actions of the rehabilitation training examples, count the occurrence frequencies of each training action in different examples, and select a preset number of training actions as basic actions based on the occurrence frequencies; perform an initial detection on the target user through the basic actions to obtain the movement parameters, physiological parameters, and environmental parameters of the target user during the completion of the basic actions, align the data according to the time stamp and perform data cleaning to obtain the preprocessed multi-source detection parameters. Integrate the health profile of the target user and the multi-source detection parameters to obtain the enhanced health profile of the target user.

[0025] Figure 3 The flowchart of constructing a training plan recommendation model in the rehabilitation training plan customization unit is shown.

[0026] According to an embodiment of the present invention, in the rehabilitation training plan customization unit, a training plan recommendation model is constructed by accessing the knowledge graph in the field of rehabilitation training. Specifically: S302. Extract the interaction information between the user's health portrait and rehabilitation training actions from rehabilitation training instances to construct an interaction graph of the user's health portrait - rehabilitation training actions, obtain the knowledge graph in the field of rehabilitation training, combine the interaction graph of the user's health portrait - rehabilitation training actions with the knowledge graph in the field of rehabilitation training to construct a training plan recommendation model, and generate a collaborative knowledge graph for rehabilitation training. S304. Use the rehabilitation training actions that interact with the enhanced health portrait of the target user to propagate in the collaborative knowledge graph, find the tail entity according to the triples in the collaborative knowledge graph, and obtain a multi-layer related entity set of the enhanced health portrait of the target user through iterative propagation. S306. Construct a triple set of the enhanced health portrait of the target user at each layer from the related entity set obtained at each layer, calculate the relevance score between all rehabilitation training actions and each triple, convert the relevance score into a probability, use the probability to perform weighted summation on the tail entity in the triple, and splice the results obtained after multiple propagations to obtain the initial representation of the enhanced health portrait of the target user. S308. Obtain a collaborative user health portrait similar to the enhanced health portrait of the target user in the collaborative knowledge graph of rehabilitation training, use a self-attention network to enhance the collaborative user health portrait, and fuse to obtain a collaborative user representation. S310. Fuse the initial representation of the enhanced health portrait of the target user with the collaborative user representation to obtain an enhanced representation of the enhanced health portrait of the target user, perform convolutional aggregation using the collaborative knowledge graph of rehabilitation training to obtain a high-order representation of rehabilitation training actions, calculate the inner product based on the enhanced representation of the enhanced health portrait of the target user and the high-order representation of rehabilitation training actions, and obtain recommended rehabilitation training actions.

[0027] It should be noted that according to the idea of collaborative filtering, the user health portraits that interact with the same rehabilitation training action often have the same "preference". Use the information of these user health portraits to describe the rehabilitation training action in more detail, that is, use other training actions of the user health portraits that have interacted with the rehabilitation training action to represent the rehabilitation training action. Combine the interaction graph of the user's health portrait - rehabilitation training actions with the knowledge graph in the field of rehabilitation training, merge the two graphs into a unified relationship graph, and generate a collaborative knowledge graph for rehabilitation training. In the collaborative knowledge graph of rehabilitation training, the user health portrait, rehabilitation training actions, and other entity nodes are connected by relationships, making full use of high-order connectivity.

[0028] Propagate the rehabilitation training actions that interact with the enhanced health profile of the target user in the collaborative knowledge graph, transfer and fuse the user's interaction information in the graph structure, more deeply mine the characteristics of the target user, and improve the accuracy of user representation. Use the rehabilitation training actions with interactions as the head entity for the first propagation, find the tail entity in the collaborative knowledge graph, and use the previous tail entity as the head entity for the next propagation to obtain the k-hop related entity set of the enhanced health profile of the target user. For each layer of triple set, the relevance of the enhanced health profile of the target user to each triple is different. Calculate the relevance scores of all rehabilitation training actions to each triple, and convert the relevance scores into probabilities through the softmax function represents the predicted rehabilitation training action among all rehabilitation training actions, represents the head entity of the triple represents the relationship between the head entity and the tail entity. Use the probability to perform a weighted sum on the tail entity in the triple, represents the user representation after k propagations of the rehabilitation training actions with interactions in the enhanced health profile of the user, represents the tail entity in the triple.

[0029] The collaborative user health profile is a group of users similar to the enhanced health profile of the target user. By leveraging the behaviors and preferences of collaborative users, potential preferences that the enhanced health profile of the target user may have are extracted. The self-attention network is used to extract important information from the collaborative user health profile for feature enhancement. Dynamically adjust the attention to its features to highlight important collaborative user health profiles, and fuse the fused collaborative user representation with the initial representation of the enhanced health profile of the target user, achieving an organic combination of collaborative information and personalized information to recommend more suitable rehabilitation training actions for the target user.

[0030] It should be noted that in the rehabilitation training plan customization unit, a personalized rehabilitation training plan for the target user is obtained. The recommended rehabilitation training actions of the target user output by the training plan recommendation model are obtained, encoded for the recommended rehabilitation training actions, and an initial rehabilitation training action combination is randomly generated, and an improved genetic algorithm is used to optimize the rehabilitation training plan; a fitness function is constructed according to the average rehabilitation effect corresponding to the rehabilitation training actions. For example, in the upper limb rehabilitation assessment of stroke, the range of motion (ROM) recovery rate and the muscle co-contraction index (CCI) are used to construct a fitness function to evaluate the rehabilitation effect. Calculate the fitness corresponding to different rehabilitation training action combinations, select a preset number of rehabilitation training action combinations according to the fitness to determine the search direction; select the rehabilitation training action combination with the highest fitness for crossover recombination during the iteration process, and change the encoding of the rehabilitation training actions during the recombination process. When the iteration times are met, output the optimal rehabilitation training action combination, and construct a personalized rehabilitation training plan for the target user based on the optimal rehabilitation training action combination combined with the physiological parameters of the target user.

[0031] It should be noted that in the feedback and control unit, multi-source detection parameters of the target user based on the personalized rehabilitation training plan are collected. From the standard action database collected by the professional medical team or sports science experiment, the standard action parameter matrix of the rehabilitation training actions in the personalized rehabilitation training is read. In the standard action parameter matrix, each row represents the key joints and body parts of an action (such as shoulder joint angle, knee joint flexion degree), and each column corresponds to the time frame of the action (standardized time axis, usually 0% - 100% of the action cycle); action features are extracted through the motion parameter sequence in the multi-source detection parameters, the continuous data stream is segmented through a sliding window or event detection to obtain independent action cycles, the rehabilitation training actions at different time stamps are identified, subsequences are labeled according to the action type (such as "elbow flexion training", "gait cycle"), a motion parameter subsequence with rehabilitation training action labels is obtained, and a motion parameter matrix of the motion parameter subsequence is constructed, and the structure of the labeled motion parameter subsequence matrix is aligned with the standard matrix; the residual matrix between the motion parameter matrix and the standard parameter matrix is calculated through element-by-element difference, and the achievable degree of the target user for each rehabilitation training action is obtained based on the residual matrix. When the achievable degree is less than the preset threshold, real-time feedback is provided through the visual interface and voice prompt to guide the target user to adjust the training action. Whether the rehabilitation training action is in an over-training state is judged according to the positive and negative values of each residual vector in the residual matrix, and relevant prompt information is generated. Preferably, when obtaining the achievable degree of the target user for each rehabilitation training action, the joint angle and motion trajectory can also be calculated through the gyroscope and acceleration, the difference information between the user's action and the standard action is compared using dynamic time warping (DTW), and the time-frequency features (such as RMS, MF, etc.) of the electromyogram signal are extracted. The muscle state is classified by machine learning random forest, and the achievable degree of each rehabilitation training action is evaluated using the difference information and muscle state. In addition, a depth camera can be set up for action capture, and micro-expressions can be further identified through a convolutional neural network. The pain level and emotional state are analyzed through facial expressions to evaluate the depressive tendency during rehabilitation.

[0032] It should be noted that in the quantitative evaluation unit, the adaptability of the personalized rehabilitation training plan is quantitatively evaluated according to the multi-source detection data and health profile of the target user during training. Retrieve rehabilitation training instances according to the health profile of the target user, set physiological parameter thresholds at different training durations based on the rehabilitation training instances, adjust the physiological parameter thresholds through the environmental parameters in the multi-source detection parameters, and use the adjusted physiological parameter thresholds to mark the abnormal points of the physiological parameters of the target user during training; obtain the rehabilitation training actions with a reachability less than the preset threshold of the target user during training and mark them, segment the training duration, and respectively obtain the density of the abnormal points of the physiological parameters and the marked rehabilitation training actions in different training duration segments; represent the adaptability score of the target user and the personalized rehabilitation training plan according to the density, preset different score threshold intervals for different training duration segments, and when the adaptability score does not meet the preset score threshold interval, it is regarded as a mismatched segment. When the number of mismatched segments reaches the preset quantity threshold, adjust the personalized rehabilitation training plan and optimize the training intensity according to the real-time performance of the patient.

[0033] Obtain the marked rehabilitation training actions with abnormal reachability of the target user, determine the human key points through the rehabilitation training parts, obtain the motion sequences of the human key points based on the standard actions of the marked rehabilitation training actions, extract the spatio-temporal features of different human key point motion sequences, construct a retrieval label through the spatio-temporal features of each human key point, use similarity calculation to obtain the rehabilitation training actions that meet the similarity standard with the retrieval label, compare the complexity of the retrieved rehabilitation training actions with the marked rehabilitation training actions, and select the rehabilitation training action with lower complexity to replace the action in the personalized rehabilitation training plan.

[0034] Figure 4 The flowchart of the intelligent rehabilitation training control method based on multi-parameter detection is shown.

[0035] The second embodiment of the present invention provides an intelligent rehabilitation training control method based on multi-parameter detection, which is applied to an intelligent rehabilitation training system based on multi-parameter detection, and includes the following steps: S402, obtain the health-related data of the target user, and real-time collect the multi-source detection parameters corresponding to the motion, physiology, and environment of the target user through preset sensors, and perform data preprocessing on the health-related data and multi-source detection parameters; S404, fuse the health profile and multi-source detection parameters of the target user, obtain the fusion features, access the knowledge graph in the rehabilitation training field to construct a training plan recommendation model, and use the fusion features as the model input to obtain the personalized rehabilitation training plan of the target user; S406. Obtain the achievability of each rehabilitation training action for the target user based on the personalized rehabilitation training plan of the target user, and provide real-time feedback through a visual interface and voice prompts according to the achievability to guide the target user to adjust the training action; S408. Quantitatively evaluate the adaptability of the personalized rehabilitation training plan based on the multi-source detection data and health profile of the target user during training, and adjust the personalized rehabilitation training plan according to the adaptability.

[0036] Retrieve and obtain rehabilitation training examples according to the health profile of the target user, set physiological parameter thresholds for different training durations based on the rehabilitation training examples, adjust the physiological parameter thresholds through the environmental parameters in the multi-source detection parameters, and use the adjusted physiological parameter thresholds to mark the physiological parameter abnormal points of the target user during training; Obtain the rehabilitation training actions with an achievability less than the preset threshold of the target user during training and mark them, segment the training duration, and respectively obtain the density of physiological parameter abnormal points and marked rehabilitation training actions in different training duration segments; Characterize the adaptability score of the target user and the personalized rehabilitation training plan according to the density, preset different score threshold intervals for different training duration segments, and when the adaptability score does not meet the preset score threshold interval, it is regarded as a mismatched segment. When the number of mismatched segments reaches the preset quantity threshold, adjust the personalized rehabilitation training plan, optimize the training intensity according to the real-time performance of the patient, and adjusting the personalized rehabilitation training plan includes action parameter-level adjustment and training plan-level adjustment. Action parameter-level adjustment includes amplitude correction and rhythm optimization. For example, if the shoulder joint flexion residual continuously > 15°, reduce the target angle by 5° - 10°, and extend the rest time between groups according to the sEMG fatigue index (such as from 30s → 45s). Training plan-level adjustment includes action replacement and difficulty reduction. For example, for patients with lumbar disc herniation, replace "forward flexion training" with "Mckenzie extension", and use an elastic band to assist in completing resistance training (the original plan was manual resistance).

[0037] The third embodiment of the present invention provides a computer-readable storage medium, which includes a program for an intelligent rehabilitation training control method based on multi-parameter detection. When the program for the intelligent rehabilitation training control method based on multi-parameter detection is executed by a processor, the steps of the intelligent rehabilitation training control method based on multi-parameter detection are implemented.

[0038] In several embodiments provided in this application, it should be understood that the disclosed method can be implemented in other ways. The embodiments described above are merely illustrative. For example, the division of the units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components can be combined, or can be integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the components shown or discussed with each other can be through some interfaces, indirect coupling or communication connection of devices or units, and can be electrical, mechanical, or other forms.

[0039] Those of ordinary skill in the art can understand that all or part of the steps of implementing the above method embodiments can be completed by hardware related to program instructions. The foregoing program can be stored in a computer-readable storage medium. When the program is executed, it executes the steps including the above method embodiments; and the foregoing storage medium includes: removable storage devices, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), magnetic disks or optical discs and other various media that can store program codes.

[0040] Alternatively, if the above integrated unit of the present invention is implemented in the form of a software functional unit and sold or used as an independent product, it can also be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the embodiments of the present invention, in essence, or the part that contributes to the prior art can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the methods described in the various embodiments of the present invention. And the foregoing storage medium includes: removable storage devices, ROM, RAM, magnetic disks or optical discs and other various media that can store program codes.

[0041] The above is only the specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention can easily think of changes or substitutions, which should all be covered within the protection scope of the present invention.

Claims

1. An intelligent rehabilitation training system based on multi-parameter detection, characterized in that The system includes a user information collection unit, a multi-source parameter collection unit, a rehabilitation training plan customization unit, a feedback and control unit, and a quantitative evaluation unit; The user information collection unit is responsible for obtaining health-related data of the target user, preprocessing the health-related data, and constructing a health profile; The multi-source parameter collection unit is responsible for collecting multi-source detection parameters corresponding to the movement, physiology, and environment of the target user in real time through preset sensors, and preprocessing the multi-source detection parameters; The rehabilitation training plan customization unit is responsible for fusing the health profile and multi-source detection parameters of the target user, obtaining an enhanced health profile of the target user, accessing the knowledge graph in the field of rehabilitation training to construct a training plan recommendation model, using the enhanced health profile as the model input, and obtaining a personalized rehabilitation training plan for the target user; The feedback and control unit is responsible for obtaining the achievable degree of each rehabilitation training action of the target user based on the personalized rehabilitation training plan of the target user, providing real-time feedback through a visual interface and voice prompts according to the achievable degree, and guiding the target user to adjust the training action; The quantitative evaluation unit is responsible for quantitatively evaluating the adaptability of the personalized rehabilitation training plan according to the multi-source detection data and health profile of the target user during training, and adjusting the personalized rehabilitation training plan according to the adaptability; 2. The intelligent rehabilitation training system based on multi-parameter detection according to claim 1, wherein In the user information collection unit, obtaining the health-related data of the target user, preprocessing the health-related data, and constructing a health profile, specifically: Obtaining medical text data as health-related data according to the identity information of the target user, performing standardization processing on the medical text data, and performing vector encoding with different granularities on the standardized medical text data using the RoBERTa pre-trained model and the Word2vec pre-trained model respectively; Obtaining the character vector and word vector corresponding to the medical text data, splicing the character vector and word vector in sequence to generate a fused vector representation, using a Bi-GRU network to perform bidirectional feature extraction on the fused vector representation, and fusing the obtained forward feature information and backward feature information to output a feature vector; Using a Bi-GRU network as the encoder module of the noise reduction encoder structure, perturbing the feature vector by adding random noise, obtaining the hidden feature distribution through linear transformation and activation function, and using the decoder module to reconstruct the hidden feature distribution to obtain a reconstructed feature vector; Fusing the reconstructed feature vector and the feature vector, and using the attention mechanism to perform position weighting representation on the fused feature vector; Obtaining label data representing health through big data methods, training a label classification layer using the label data, performing classification prediction on the position-weighted fused feature vector, obtaining labels that meet the probability threshold, and constructing a health profile of the target user based on the obtained labels; 3. The intelligent rehabilitation training system based on multi-parameter detection according to claim 1, wherein, In the multi-source parameter collection unit, collecting multi-source detection parameters corresponding to the movement, physiology, and environment of the target user in real time through preset sensors, specifically: Retrieve and obtain rehabilitation training instances according to the health profile of the target user, decompose the training actions of the rehabilitation training instances, count the occurrence frequencies of each training action in different instances, and select a preset number of training actions as basic actions based on the occurrence frequencies; Conduct an initial detection of the target user through the basic actions, obtain the motion parameters, physiological parameters, and environmental parameters of the target user during the completion of the basic actions, align the data according to the time stamps and perform data cleaning to obtain the preprocessed multi-source detection parameters.

4. The intelligent rehabilitation training system based on multi-parameter detection according to claim 1, characterized in that, In the rehabilitation training plan customization unit, connect to the knowledge graph in the field of rehabilitation training to construct a training plan recommendation model. Specifically: Extract the interaction information between the user health profile and rehabilitation training actions from the rehabilitation training instances to construct an interaction graph of user health profile - rehabilitation training actions, obtain the knowledge graph in the field of rehabilitation training, combine the interaction graph of user health profile - rehabilitation training actions with the knowledge graph in the field of rehabilitation training to construct a training plan recommendation model, and generate a collaborative knowledge graph for rehabilitation training; Use the rehabilitation training actions that interact with the enhanced health profile of the target user to propagate in the collaborative knowledge graph, find the tail entity according to the triples in the collaborative knowledge graph, and obtain the multi-layer related entity set of the enhanced health profile of the target user through iterative propagation; Construct a triple set of the enhanced health profile of the target user at each layer for the related entity set obtained at each layer, calculate the relevance score between all rehabilitation training actions and each triple, convert the relevance score into a probability, use the probability to perform weighted summation on the tail entity in the triple, and splice the results obtained after multiple propagations to obtain the initial representation of the enhanced health profile of the target user; Obtain a collaborative user health profile similar to the enhanced health profile of the target user in the collaborative knowledge graph of rehabilitation training, use a self-attention network to enhance the collaborative user health profile, and fuse to obtain the collaborative user representation; Fuse the initial representation of the enhanced health profile of the target user with the collaborative user representation to obtain the enhanced representation of the enhanced health profile of the target user, and use the collaborative knowledge graph of rehabilitation training for convolutional aggregation to obtain the high-order representation of the rehabilitation training actions. Calculate the inner product according to the enhanced representation of the enhanced health profile of the target user and the high-order representation of the rehabilitation training actions to obtain the recommended rehabilitation training actions.

5. The intelligent rehabilitation training system based on multi-parameter detection according to claim 4, characterized in that, In the rehabilitation training plan customization unit, obtain the personalized rehabilitation training plan of the target user. Specifically: Obtain the recommended rehabilitation training actions of the target user output by the constructed training plan recommendation model, encode the recommended rehabilitation training actions, randomly generate an initial rehabilitation training action combination, and use an improved genetic algorithm to optimize the rehabilitation training plan; Construct a fitness function according to the average rehabilitation effect corresponding to the rehabilitation training actions, calculate the fitness corresponding to different rehabilitation training action combinations, and select a preset number of rehabilitation training action combinations for replication according to the fitness to determine the search direction; During the iteration process, select the rehabilitation training action combination with the highest fitness for crossover recombination, and change the encoding of the rehabilitation training actions during the recombination process. When the iteration times are met, output the optimal rehabilitation training action combination, and construct a personalized rehabilitation training plan for the target user based on the optimal rehabilitation training action combination combined with the physiological parameters of the target user.

6. The intelligent rehabilitation training system based on multi-parameter detection according to claim 1, wherein, In the feedback and control unit, obtain the achievability of each rehabilitation training action for the target user based on the personalized rehabilitation training plan for the target user. Specifically: Collect multi-source detection parameters of the target user based on the personalized rehabilitation training plan, and read the standard action parameter matrix of the rehabilitation training actions in the personalized rehabilitation training; Extract action features through the motion parameter sequence in the multi-source detection parameters, identify the rehabilitation training actions at different timestamps, obtain the motion parameter subsequences with rehabilitation training action labels, and construct the motion parameter matrix of the motion parameter subsequences; Calculate the residual matrix of the motion parameter matrix and the standard parameter matrix, obtain the achievability of each rehabilitation training action for the target user based on the residual matrix. When the achievability is less than the preset threshold, provide real-time feedback through the visualization interface and voice prompt to guide the target user to adjust the training actions.

7. The intelligent rehabilitation training system based on multi-parameter detection according to claim 1, characterized in that, In the quantitative evaluation unit, quantitatively evaluate the adaptability of the personalized rehabilitation training plan according to the multi-source detection data and health portrait of the target user during training. Specifically: Retrieve rehabilitation training instances according to the health portrait of the target user, set physiological parameter thresholds under different training durations based on the rehabilitation training instances, adjust the physiological parameter thresholds through the environmental parameters in the multi-source detection parameters, and use the adjusted physiological parameter thresholds to mark the physiological parameter abnormal points of the target user during training; Obtain and mark the rehabilitation training actions with achievability less than the preset threshold of the target user during training, segment the training duration, and respectively obtain the density of physiological parameter abnormal points and the marked rehabilitation training actions in different training duration segments; Characterize the adaptability score of the target user and the personalized rehabilitation training plan according to the density, preset different score threshold intervals for different training duration segments. When the adaptability score does not meet the preset score threshold interval, it is regarded as a mismatched segment. When the number of mismatched segments reaches the preset quantity threshold, adjust the personalized rehabilitation training plan.

8. An intelligent rehabilitation training control method based on multi-parameter detection, characterized in that, Applied to the intelligent rehabilitation training system based on multi-parameter detection as described in any one of claims 1-7, including the following steps: Obtain the health-related data of the target user, and collect the multi-source detection parameters corresponding to the motion, physiology, and environment of the target user in real time through preset sensors, and perform data preprocessing on the health-related data and multi-source detection parameters; Fuse the health portrait and multi-source detection parameters of the target user, obtain the fusion features, access the knowledge graph in the field of rehabilitation training to construct a training plan recommendation model, and use the fusion features as the model input to obtain the personalized rehabilitation training plan for the target user; Based on the personalized rehabilitation training plan for the target user, obtain the achievable degree of the target user for each rehabilitation training action, and provide real-time feedback through a visual interface and voice prompts according to the achievable degree to guide the target user to adjust the training action; Quantify and evaluate the adaptability of the personalized rehabilitation training plan according to the multi-source detection data and health profile of the target user during training, and adjust the personalized rehabilitation training plan according to the adaptability.

Images