Intelligent remote rehabilitation guidance system

Abstract

The invention belongs to the technical field of remote rehabilitation guidance, and particularly relates to an intelligent remote rehabilitation guidance system. The system integrates the functions of man-machine interaction, information encryption, physiological and psychological assessment, rehabilitation grade division, physician professional assessment, rehabilitation knowledge base construction, rehabilitation matching planning, rehabilitation training monitoring and feedback, rehabilitation effect prediction and the like, and a fuzzy assessment algorithm and an emotional depression detection algorithm are integrated. The system comprehensively evaluates the physiological and psychological health conditions of a rehabilitative person, recommends appropriate rehabilitation doctors or health planning guidance suggestions for the rehabilitative person, monitors the standard degree and micro-expression changes of training actions in real time and gives timely feedback and adjustment suggestions in the rehabilitation process, and meanwhile, predicts a rehabilitation track by utilizing a machine learning algorithm, so that the rehabilitation effect is improved. Subsequent rehabilitation planning is optimized, and personalized precise rehabilitation is ensured; according to the invention, the participation degree and training scientificity of the convalescent are improved, and an effective solution is provided for remote rehabilitation guidance.

Description

Technical Field

[0001] The invention belongs to the technical field of remote rehabilitation guidance, and in particular discloses an intelligent remote rehabilitation guidance system. Background Art

[0002] With the rapid development of information technology, remote rehabilitation guidance systems, as an emerging means of health management, are gradually changing the traditional rehabilitation medical model. Such systems are usually based on Internet technology. Through video communication, mobile applications, wearable devices and other technical means, rehabilitation treatment is no longer restricted by geographical location, and can provide remote, personalized rehabilitation training plans and real-time guidance for patients. It includes functions such as patient information management, rehabilitation plan formulation, remote monitoring, and data analysis, aiming to improve rehabilitation efficiency, reduce rehabilitation costs, and allow patients to enjoy professional rehabilitation services at home.

[0003] For example, the patent with the authorization announcement number CN114758747B discloses a remote rehabilitation guidance system and method based on big data. In the early stage, different instructors and rehabilitated persons are respectively profiled, so that instructors with different abilities can perform portrait analysis on rehabilitated persons in different states to obtain guidance coefficients and rehabilitation coefficients. Based on the guidance coefficients, the guidance effects of the instructors can be evaluated and graded, so that instructors of different levels can provide corresponding rehabilitation training for rehabilitated persons with different degrees of severity. Based on the rehabilitation coefficients, the initial states of rehabilitated persons during training can be evaluated and graded, so that rehabilitated persons of different levels can be matched with instructors with different guidance abilities.

[0004] For example, the patent with publication number CN115527665A discloses a method of a medical rehabilitation remote guidance system based on augmented reality and semantic control, which includes the following steps: S1: The medical staff wears augmented reality glasses, and interacts with the nursing assistance robot built into the glasses according to the glasses interface and voice prompts to further understand the relevant information of the rehabilitated person; S2: The medical staff initiates a remote consultation and selects a rehabilitation physician or nursing expert online in the background for video communication. The background expert can directly view the picture captured by the augmented reality glasses (i.e., the picture directly seen by the medical staff) and guide the medical staff. In addition, the expert can also send text or pictures, which will be directly displayed on the screen of the augmented reality glasses for the convenience of medical staff to view.

[0005] The above existing technologies have the following problems: 1) The existing technologies often only provide physiological testing guidance when guiding the rehabilitation of the rehabilitated persons, and are unable to capture and guide the changes in the mental health of the rehabilitated persons in real time; 2) Most of the existing technical rehabilitation guidance is provided by a single department or physician, and for the sudden conditions of the rehabilitated persons, it is impossible to quickly switch health instructors or provide joint guidance across departments; 3) In addition, the monitoring of some abnormal conditions that occur during the training process is also easily overlooked by the existing technologies. In order to solve the above problems, the present invention provides an intelligent remote rehabilitation guidance system. Summary of the invention

[0006] In view of the shortcomings of the prior art, the present invention proposes an intelligent remote rehabilitation guidance system, which integrates human-computer interaction, information encryption, physiological and psychological assessment, rehabilitation level classification, physician professional assessment, rehabilitation knowledge base construction, rehabilitation matching planning, rehabilitation training monitoring and feedback, and rehabilitation effect prediction. Through the comprehensive fuzzy evaluation algorithm and emotional depression detection algorithm, the system comprehensively evaluates the physiological and mental health of the rehabilitated person, and recommends suitable rehabilitation physicians or health planning guidance opinions. During the rehabilitation process, the system monitors the standardization of training movements and micro-expression changes in real time, and gives timely feedback and adjustment suggestions. At the same time, it uses machine learning algorithms to predict rehabilitation trajectories, optimize subsequent rehabilitation plans, and ensure personalized and accurate rehabilitation. The present invention improves the participation of rehabilitated people and the scientific nature of training, and provides an effective solution for remote rehabilitation guidance.

[0007] To achieve the above object, the present invention provides the following technical solutions:

[0008] An intelligent remote rehabilitation guidance system, comprising: a rehabilitation patient portrait module, a rehabilitation planning module and a remote interactive monitoring module;

[0009] The rehabilitation patient portrait module includes a physiological assessment unit, a psychological assessment unit, and a rehabilitation level portrait unit;

[0010] The physiological assessment unit is used to obtain the first health data of the rehabilitated person, and calculate the first health score of the rehabilitated person through the configured comprehensive fuzzy assessment algorithm; the psychological assessment unit is used to obtain the second rehabilitation data of the rehabilitated person, and detect and calculate the second health score of the rehabilitated person through the configured psychological questionnaire and depression detection algorithm; the rehabilitation level portrait unit is used to calculate the comprehensive rehabilitation score of the rehabilitated person according to the calculated first health score and second health score, and classify the rehabilitation guidance level of the rehabilitated person according to the obtained comprehensive rehabilitation score;

[0011] The rehabilitation planning module includes the physician portrait unit, rehabilitation knowledge base unit and rehabilitation matching planning unit;

[0012] The physician portrait unit is used to obtain the physician's professional level and work results data, call the comprehensive fuzzy evaluation algorithm in the physiological evaluation unit, and calculate the rehabilitation physician's professional score and coachability level; the rehabilitation knowledge base unit is used to obtain the historical rehabilitation case library data, build a rehabilitation knowledge base through the configured natural language algorithm, and build a built-in reasoning model for the constructed rehabilitation knowledge base to obtain a rehabilitation reasoning knowledge base with reasoning ability; the rehabilitation matching planning unit is used to recommend a matching rehabilitation physician across departments for the rehabilitated person through the matching recommendation algorithm based on the obtained comprehensive rehabilitation score of the rehabilitated person, or recommend a matching rehabilitation knowledge base to the rehabilitated person through the matching recommendation algorithm. Online health guidance opinions or guidance videos generated by the intelligent generation algorithm;

[0013] The remote interactive monitoring module includes a motion monitoring unit and a micro-expression monitoring unit; the motion monitoring unit is used to monitor the standard of the rehabilitation person's training movements in real time according to the configured training motion evaluation algorithm, and to provide scoring feedback on the standard of the training movements; the micro-expression monitoring unit is used to monitor the changes in the micro-expressions of the rehabilitation person during the training process in real time according to the configured micro-expression monitoring algorithm, and to provide feedback on the monitoring results when the monitored micro-expressions change abnormally.

[0014] Specifically, the specific steps of the physiological assessment unit calculating the first health score of the recovering person include:

[0015] A1. Use various physiological monitoring devices to collect the real-time first health data of the rehabilitated person, clean the collected data, remove abnormal values, standardize the preprocessing, and obtain the first health data after preprocessing;

[0016] A2. Constructing a first-level evaluation factor set s=[s1…s l …s L ], and use the sub-factors corresponding to the first health factor set to construct the secondary evaluation factor set s l =[s l1 …s lk …sl K ], sl represents the lth first-level evaluation factor in the first-level evaluation factor set, s lk represents the kth secondary evaluation factor in the lth primary evaluation factor; L represents the length of the primary evaluation factor set, and K represents the length of the secondary evaluation factor set corresponding to the lth primary evaluation factor;

[0017] A3. Set the rehabilitation index level set to v = [v1…v m …v M ], evaluate and score the constructed first-level evaluation factor set and second-level evaluation factor set by expert scoring, and obtain the fuzzy evaluation matrix R of the first-level evaluation factor set and the fuzzy evaluation matrix R' of the second-level evaluation factor set in turn; where v mrepresents the mth rehabilitation index level, and M represents the total number of rehabilitation index levels.

[0018] Specifically, the physiological assessment unit calculates the first health score of the recovering person in the following specific steps:

[0019] A4. According to the obtained R', calculate the weight W of the second-level evaluation factor under the lth first-level evaluation factor through the minimum membership weighted average deviation l =[w l1 …w lk …w lK ], and calculate the comprehensive evaluation vector C of K secondary evaluation factors l =[C l1 …C lm …C lM ], as follows: where w lk represents the weight of the kth secondary evaluation factor under the lth primary evaluation factor, c lm represents the comprehensive evaluation value of the secondary evaluation factor under the lth primary evaluation factor at the mth rehabilitation indicator level, gk represents the maximum membership of the kth secondary evaluation factor and the rehabilitation indicator level set; r' km It indicates the membership degree of the k-th secondary assessment factor to the m-th rehabilitation indicator level;

[0020] A5. Based on the secondary assessment factors l The comprehensive evaluation vector C of the evaluation index level l , calculate the first-level evaluation factor index weight W s , by obtaining W s =(w1…w l …w L ), using the fuzzy evaluation matrix R, W of the first-level evaluation factor set s The comprehensive evaluation vector of the first-level evaluation factor is calculated by the comprehensive evaluation result formula of a single first-level evaluation factor

[0021] A6. Comprehensively evaluate the calculated first-level evaluation factors Add up all the elements to get the first health score of the recovered person.

[0022] Specifically, the specific workflow steps of the rehabilitation matching planning unit include:

[0023] E1. Set a rehabilitation score selection threshold. When the comprehensive rehabilitation score obtained is lower than the rehabilitation score selection threshold, two selection buttons, namely rehabilitation physician and rehabilitation reasoning knowledge base, are provided through the guidance system interactive interface for the rehabilitation patient to choose.

[0024] E2. When the patient selects a rehabilitation video generated by the rehabilitation reasoning knowledge base, the corresponding rehabilitation guidance video and rehabilitation guidance suggestions are recommended to the patient through a matching recommendation algorithm based on the obtained comprehensive rehabilitation score, the first health score, and the second health score of the patient;

[0025] E3. When the rehabilitation reasoning knowledge base does not contain rehabilitation videos and rehabilitation suggestions corresponding to the comprehensive rehabilitation score, first health score and second health score of the rehabilitated person, the reasoning model is called to generate corresponding rehabilitation guidance videos and rehabilitation guidance suggestions for the rehabilitated person through the existing rehabilitation knowledge base using the video generation algorithm and the text generation algorithm.

[0026] Specifically, the specific workflow steps of the rehabilitation matching planning unit also include:

[0027] E4. When the patient selects a rehabilitation physician, a matching threshold is set, and the matching coefficient between the patient and the rehabilitation physician is calculated through a matching recommendation algorithm based on the patient's comprehensive rehabilitation score, the first health score and the second health score, and the rehabilitation physician's professional score and coachability level. When the matching coefficient is greater than the matching threshold, the corresponding rehabilitation physician is recommended. If it is less than the matching coefficient, the matching continues until the matching is completed. If the rehabilitation physician whose matching coefficient is greater than the matching threshold cannot be matched after the matching is completed, the rehabilitation physician with the largest matching coefficient with the patient is recommended to the corresponding patient;

[0028] E5. Set a mental health threshold. When the second health score is greater than the mental health threshold, in addition to the rehabilitation physician recommended by E4, the corresponding mental health physician is recommended to the rehabilitated person across departments through a matching recommendation algorithm based on the second health score. If it is less than the threshold, no mental health physician is recommended.

[0029] Specifically, the specific workflow steps of the rehabilitation matching planning unit also include:

[0030] E6. When the comprehensive rehabilitation score obtained is higher than the rehabilitation score selection threshold, the matching recommendation algorithm is used to directly recommend the corresponding rehabilitation physician to the patient, and at the same time, the matching recommendation algorithm is used to recommend a mental health physician to the patient across departments based on the second health score;

[0031] E7. When the comprehensive rehabilitation score of the patient is lower than the rehabilitation score selection threshold after training guided by the rehabilitation physician and mental health physician recommended by E6, the patient will undergo the E1-E5 matching recommendation process.

[0032] Specifically, the specific steps of the action monitoring unit scoring the standard of the training action include:

[0033] F1. Obtain rehabilitation physician video rehabilitation guidance data, standardized rehabilitation guidance rule text data and historical rehabilitation training video data, and perform denoising preprocessing on the collected data. At the same time, perform frame extraction on the preprocessed rehabilitation physician video rehabilitation guidance data and historical rehabilitation training video data through the OPenCV library to obtain rehabilitation training time series image data;

[0034] F2. Construct a training action evaluation model based on the Chinese pre-trained Bert sub-model and the pre-trained HRNet, and input the acquired rehabilitation training time series image data into the HRNet in the training action evaluation model to obtain the training action joint node height and curvature features. At the same time, input the pre-processed standardized rehabilitation guidance rule text data into the Chinese pre-trained Bert sub-model to extract the standard training rule threshold features, and input the acquired training action joint node height and curvature features and the acquired standard training rule threshold features into the cross-attention sub-model for fusion operation to obtain the fusion training standard features.

[0035] F3. Input the acquired fusion training standard features into the output layer of the training action evaluation model to output the corresponding training action score. At the same time, use the extracted training action joint node height and curvature features and the standard training rule threshold features to calculate the training standard error. Use the training standard error to construct a cross entropy function, train the training action evaluation model, and obtain the completed training action evaluation model.

[0036] Specifically, the specific steps of the action monitoring unit scoring the standard of the training action also include:

[0037] F4. Deploy the training action evaluation model after training to the action monitoring unit. When the rehabilitated person is trained by the rehabilitation physician or rehabilitation video recommended by the guidance system, the deployed sensors collect the training actions of the rehabilitated person in real time, and upload the collected action data to the action monitoring unit for real-time scoring of the training action standard.

[0038] F5. Set the training action standard score threshold. When the obtained real-time training action standard score is less than the training action standard score threshold, feedback is given and the training action is adjusted in real time.

[0039] Specifically, the specific workflow of the micro-expression monitoring unit includes:

[0040] G1. Obtain an annotated public micro-expression database and historical training video data collected during rehabilitation training, use a pre-trained face detection model to extract and annotate the face area from the collected historical training video data, and integrate it with the obtained public micro-expression database;

[0041] G2, perform key point labeling and image alignment on the unified and integrated micro-expression data, and perform filtering and enhancement processing on the aligned images to obtain a micro-expression training data set;

[0042] G3, construct a micro-expression detection model, input the images in the acquired micro-expression training data set into the initial feature extraction layer of the micro-expression detection model in order of time points, and obtain the key features of facial expressions;

[0043] G4, input the key features of the acquired facial expression into the optical flow sub-model layer in the micro-expression detection model to obtain the dynamic change features of the local micro-expressions in the key features of the facial expression;

[0044] G5, input the key features of facial expressions and the dynamic change features of local micro-expressions into the collaborative attention layer, perform feature fusion, and obtain fused micro-expression features;

[0045] G6. Input the acquired fused micro-expression features into the classification layer in the model, output the probability of detecting the corresponding type of micro-expression, set the cross entropy function as the model loss function, and set the training cycle to 200 to train the micro-expression detection model to obtain the trained micro-expression detection model.

[0046] Specifically, the specific workflow of the micro-expression monitoring unit also includes:

[0047] G7. Deploy the acquired micro-expression detection model to the micro-expression monitoring unit. When the rehabilitated person is training, the sensor collects the facial image or video data of the rehabilitated person in real time, transmits the collected data to the micro-expression monitoring unit for detection, and outputs the micro-expression detection results.

[0048] G8. Set an abnormal probability threshold. When the probability that the detected output micro-expression is a negative micro-expression is greater than the abnormal probability threshold, an alarm feedback is performed, otherwise no alarm feedback is performed.

[0049] Compared with the prior art, the present invention has the following beneficial effects:

[0050] 1. The present invention introduces a meticulous real-time monitoring mechanism through motion monitoring and micro-expression monitoring units to dynamically track the movement standards and emotional changes during rehabilitation training. This instant feedback mechanism can not only effectively correct non-standard movements and prevent training injuries, but also provide personalized encouragement and adjustment suggestions through the intelligent rehabilitation assistant based on the emotional response of the rehabilitated person, thereby enhancing the rehabilitated person's enthusiasm and sense of participation. At the same time, this mechanism also promotes real-time communication between doctors and rehabilitated people, making the adjustment of the training process more scientific and timely.

[0051] 2. The present invention aims at the limitation of traditional rehabilitation technology that ignores mental health. Through the psychological assessment unit and the emotion and depression detection algorithm, the mental health changes of the rehabilitated person are captured in real time, ensuring that both physical and mental health are given equal importance, and realizing comprehensive rehabilitation guidance. This two-pronged approach can more comprehensively evaluate the progress of rehabilitation, timely adjust the rehabilitation strategy, effectively respond to the psychological fluctuations of the rehabilitated person, and promote the synchronous recovery of body and mind.

[0052] 3. In the face of the limitation of a single guidance resource, the present invention not only improves the accuracy of professional evaluation and matching of rehabilitation physicians through the physician portrait unit and the rehabilitation knowledge base unit, but also realizes rapid rehabilitation guidance adjustment and cross-department joint guidance capabilities through the built-in reasoning algorithm and interdisciplinary rehabilitation reasoning knowledge base. When a patient encounters an emergency, the system can quickly identify and recommend the most appropriate rehabilitation resource intervention, including replacing rehabilitation physicians or initiating cross-department collaboration plans, which greatly shortens the emergency response time and improves rehabilitation efficiency and safety. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 This is a module diagram of an intelligent remote rehabilitation guidance system according to Embodiment 1 of the present invention;

[0054] Figure 2 This is a structural diagram of a depression detection and evaluation model according to Example 1 of the present invention;

[0055] Figure 3 This is a workflow diagram of an intelligent remote rehabilitation guidance system according to Embodiment 2 of the present invention. DETAILED DESCRIPTION

[0056] Example 1

[0057] See also Figure 1 , an embodiment provided by the present invention: an intelligent remote rehabilitation guidance system, including a human-computer interaction module, a rehabilitator portrait module, a rehabilitation planning module, a remote interactive monitoring module, a rehabilitation assessment and suggestion module and an encryption module;

[0058] The human-computer interaction module is used for the login of the rehabilitated person and the rehabilitation physician, filling in the basic information of the rehabilitated person and the rehabilitation physician, and uploading the health status information of the rehabilitated person and the professional level data of the rehabilitation physician;

[0059] In this embodiment, the health status information includes physiological health information and psychological health information; the professional level data of rehabilitation physicians include professional ability data and work achievement data of rehabilitation physicians; the professional ability data include academic level, professional knowledge level and corresponding ability certification certificates; the work achievement data include the length of time for rehabilitation of the guided rehabilitation personnel in the historical rehabilitation guidance process, the specific number of people who have recovered after guidance and the relapse rate after recovery; for example, the proportion of depression patients who have relapsed after guidance rehabilitation;

[0060] The rehabilitated patient portrait module is used to collect the physiological and psychological data of the rehabilitated patient and construct a rehabilitated patient portrait through a machine learning algorithm; the rehabilitated patient portrait module includes a physiological assessment unit, a psychological assessment unit and a rehabilitation level portrait unit;

[0061] The physiological assessment unit is used to obtain the first health data of the rehabilitated person, and calculate the first health score of the rehabilitated person through the configured comprehensive fuzzy assessment algorithm; the psychological assessment unit is used to obtain the second rehabilitation data of the rehabilitated person, and detect and calculate the second health score of the rehabilitated person through the configured psychological questionnaire and depression detection algorithm; the rehabilitation level portrait unit is used to calculate the comprehensive rehabilitation score of the rehabilitated person according to the calculated first health score and second health score, and divide the rehabilitated person into rehabilitation guidance levels according to the obtained comprehensive rehabilitation score; further, this embodiment divides the rehabilitation guidance level into 5 levels, namely excellent rehabilitation level, good rehabilitation level, medium rehabilitation level, rehabilitation level that needs to be strengthened and low rehabilitation level; among which the comprehensive rehabilitation score range of the excellent rehabilitation level is: 0 points to 4 points, the rehabilitated person shows an extremely high level of rehabilitation, the physiological function is close to or has recovered to the pre-illness state, the psychological state is good, the self-management ability is strong, the compliance with the rehabilitation plan is high, and there is obvious rehabilitation progress and maintenance ability; the comprehensive rehabilitation of the good rehabilitation level Scoring range: 5 to 8 points. The patient has made significant progress in rehabilitation, his physiological function and daily activity ability have been greatly improved, his mental health is stable, and he can effectively participate in and benefit from rehabilitation activities, but there is still room for improvement in some aspects; the comprehensive rehabilitation score range of medium rehabilitation level: 9 to 12 points. The patient has made obvious progress in rehabilitation, but there are some persistent physiological or psychological disorders, and more targeted rehabilitation interventions and support are needed, and he may have certain difficulties in some daily activities; the comprehensive rehabilitation score range of the rehabilitation level that needs to be strengthened: 13 to 15 points. The patient's rehabilitation process is slow, and there are more physiological or psychological dysfunctions. Strengthened rehabilitation measures are needed, including but not limited to more intensive rehabilitation training, psychological counseling or the use of assistive devices to promote further rehabilitation; the comprehensive rehabilitation score range of low rehabilitation level: 16 to 20 points. The patient's rehabilitation progress is slow or stagnant, and there are serious physiological limitations or psychological problems. He needs high attention and personalized critical rehabilitation management, which may involve collaborative treatment by a multidisciplinary team to prevent deterioration and explore all possible rehabilitation pathways. Furthermore, in this embodiment, the first health score is the physiological health score of the rehabilitated person, and the second health score is the psychological health score of the rehabilitated person; the first health data is the physiological health parameter, and the second rehabilitation data is the psychological health parameter.

[0062] The rehabilitation planning module is used to construct a rehabilitation guidance planning library, and recommend a matching rehabilitation plan to the rehabilitated person through a recommendation algorithm based on the obtained comprehensive rehabilitation score and rehabilitation level; the rehabilitation planning module includes a physician portrait unit, a rehabilitation knowledge base unit and a rehabilitation matching planning unit; the physician portrait unit is used to obtain the physician's professional level and work performance data, call the comprehensive fuzzy evaluation algorithm in the physiological evaluation unit, and calculate the rehabilitation physician's professional score and guide level; the rehabilitation knowledge base unit is used to obtain historical rehabilitation case library data, build a rehabilitation knowledge base through a configured natural language algorithm, and build a built-in reasoning model for the constructed rehabilitation knowledge base to obtain a rehabilitation reasoning knowledge base with reasoning ability; the rehabilitation matching planning unit is used to recommend a matching rehabilitation physician across departments to the rehabilitated person through a matching recommendation algorithm based on the obtained comprehensive rehabilitation score of the rehabilitated person, or recommend a matching rehabilitation reasoning knowledge base to the rehabilitated person through an online health guidance opinion or guidance video generated by an intelligent generation algorithm; the matching recommendation algorithms in this embodiment include BERT4Rec, SimBERT and ERNIE-Gram;

[0063] The remote interactive monitoring module is used to collect and monitor abnormal data of movements and micro-expressions during the rehabilitation process, and adjust the rehabilitation process through the configured intelligent rehabilitation assistant according to the abnormal data; the remote interactive monitoring module includes a movement monitoring unit, a micro-expression monitoring unit and an intelligent interaction unit; the movement monitoring unit is used to monitor the standard of the rehabilitation person's training movements in real time according to the configured training movement evaluation algorithm, and to score and feedback the standard of the training movements; the micro-expression monitoring unit is used to monitor the changes in the micro-expressions of the rehabilitation person during the training process in real time according to the configured micro-expression monitoring algorithm, and when the monitored micro-expressions change abnormally, the monitoring results are fed back; the intelligent interaction unit is used to give the rehabilitation person corresponding feedback, encouragement and adjustment suggestions according to the training movement standard score and micro-expression changes monitored by the rehabilitation person during the rehabilitation training process through the configuration of the intelligent rehabilitation assistant, and feedback the non-standard training movements or training abnormalities that occur during the training process to the corresponding recommended matching rehabilitation physician, so as to adjust the training movements and training process in real time and improve the participation and compliance of rehabilitation; further, the intelligent rehabilitation assistant in this embodiment is implemented by configuring an intelligent voice robot;

[0064] The rehabilitation assessment and suggestion module is used to analyze the rehabilitation trend according to the acquired rehabilitation process data of the rehabilitated person, and adjust the recommended rehabilitation plan according to the analysis results; the rehabilitation assessment module includes a rehabilitation assessment unit and a rehabilitation suggestion and feedback unit; the rehabilitation assessment unit is used to obtain the data obtained during each rehabilitation guidance process of the rehabilitated person, and use the integrated machine learning algorithm to predict and analyze the obtained data to obtain the comprehensive rehabilitation score change trajectory of the rehabilitated person; the machine learning algorithm in this embodiment includes a support vector machine, a multivariate regression algorithm, and a random deep forest algorithm;

[0065] The rehabilitation suggestion and feedback unit is used to feed back the acquired comprehensive rehabilitation score change trajectory of the rehabilitated person to the rehabilitation matching planning unit, and adjust the subsequent recommended rehabilitation planning process; for example, adjust the recommended physician, rehabilitation suggestion, rehabilitation video or rehabilitation guidance cycle, and make corresponding adjustments to the suggestions on lifestyle and eating habits to promote continuous rehabilitation;

[0066] The encryption module is used to encrypt and protect the basic information of the rehabilitated persons and rehabilitation physicians, as well as the health status information of the rehabilitated persons and the professional level data of the rehabilitation physicians through an encryption processing algorithm; the encryption processing algorithm includes a symmetric key encryption algorithm, a hash algorithm and a key management algorithm.

[0067] Furthermore, in this embodiment, the specific steps of the physiological assessment unit calculating the first health score of the recovered person include:

[0068] A1. Use various physiological monitoring equipment to collect the real-time primary health data of the rehabilitated person, clean the collected data, remove abnormal values, standardize and pre-process, and obtain the primary health data after pre-processing; physiological monitoring equipment includes heart rate monitors, blood pressure monitors, blood glucose meters, activity monitoring bracelets, etc.; physiological parameters include cardiovascular function data, musculoskeletal function, metabolism and nutrition data, and activity capacity data;

[0069] A2. Constructing a first-level evaluation factor set s=[s1…s l …s L ], and use the sub-factors corresponding to the first health factor set to construct the secondary evaluation factor set s l =[s l1 …s lk …s lK ],s l represents the lth first-level evaluation factor in the first-level evaluation factor set, s lk represents the kth secondary evaluation factor in the lth primary evaluation factor; L represents the length of the primary evaluation factor set, and K represents the length of the secondary evaluation factor set corresponding to the lth primary evaluation factor;

[0070] In this embodiment, the first health data includes cardiovascular function indicators, musculoskeletal function indicators, metabolic and nutritional indicators, and activity indicators; cardiovascular function indicators include heart rate variability sub-factors, resting heart rate sub-factors, and maximum oxygen uptake quantum factors; musculoskeletal function indicators include muscle strength and endurance sub-factors, joint range of motion sub-factors, and balance ability sub-factors; metabolic and nutritional indicators include weight and body composition sub-factors and blood sugar sub-factors; activity indicators include walking ability sub-factors.

[0071] A3. Set the rehabilitation index level set to v = [v1…v m …v M ], evaluate and score the constructed first-level evaluation factor set and second-level evaluation factor set by expert scoring, and obtain the fuzzy evaluation matrix R of the first-level evaluation factor set and the fuzzy evaluation matrix R' of the second-level evaluation factor set in turn; where v m represents the mth rehabilitation index level, and M represents the total number of rehabilitation index levels; in this embodiment, the rehabilitation index levels include five levels: Ⅰ, Ⅱ, Ⅲ, Ⅳ, and Ⅴ, among which Ⅰ corresponds to 8-10 points, Ⅱ corresponds to 6-8 points, Ⅲ corresponds to 4-6 points, Ⅳ corresponds to 2-4 points, and Ⅴ corresponds to 0-2 points. The lower the score obtained, the more serious the corresponding physiological state of the rehabilitator is, and the greater the difficulty of rehabilitation.

[0072] A4. According to the obtained R', calculate the weight W of the second-level evaluation factor under the lth first-level evaluation factor through the minimum membership weighted average deviation l =[w l1 …w lk …w lK ], and calculate the comprehensive evaluation vector C of K secondary evaluation factors l =[C l1 …C lm …C lM ], as follows: where w lk represents the weight of the kth secondary evaluation factor under the lth primary evaluation factor; c lm represents the comprehensive evaluation value of the second-level evaluation factor under the lth first-level evaluation factor at the mth rehabilitation index level, g k represents the maximum membership of the kth secondary assessment factor and the rehabilitation indicator level set; r' km Indicates the membership degree of the kth secondary assessment factor to the mth rehabilitation indicator level; the r' km The value is obtained through R';

[0073] A5. Based on the secondary assessment factors l The comprehensive evaluation vector C of the evaluation index level l , calculate the first-level evaluation factor index weight W s , by obtaining Ws =(w1…w l …w L ), using the fuzzy evaluation matrix R, W of the first-level evaluation factor set s The comprehensive evaluation vector of the first-level evaluation factor is calculated by the comprehensive evaluation result formula of a single first-level evaluation factor Specifically, the comprehensive evaluation result formula of a single first-level evaluation factor is: Among them, r lm represents the degree of membership of the lth primary assessment factor to the mth rehabilitation indicator level, s l represents the minimum value of the membership of the lth first-level assessment factor and all rehabilitation indicator levels in the rehabilitation indicator level set, w l represents the indicator weight of the lth first-level evaluation factor, c m It represents the comprehensive evaluation value of the first-level evaluation factor at the mth rehabilitation indicator level;

[0074] A6. Comprehensively evaluate the calculated first-level evaluation factors Add up all the elements to get the first health score of the recovered person.

[0075] Furthermore, the psychological assessment unit in this implementation provides a second health score for the rehabilitated person through a two-faceted process;

[0076] First, by building a psychological assessment questionnaire into the remote guidance system, before the user starts training, the questionnaire report is pushed to the patient through the configured intelligent rehabilitation assistant, and the patient fills in the psychological assessment questionnaire. After the questionnaire is completed, the assessment algorithm built into the guidance system is used to evaluate the patient's data obtained from the psychological assessment questionnaire to obtain the first psychological score. The questionnaires used in the questionnaire survey include the Patient Health Questionnaire-Depression Module and the Hamilton Depression Rating Scale.

[0077] Secondly, through the configured depression detection and evaluation model, the facial image data of the recovered persons is captured in real time by a visual device, and a depression detection score is given to the recovered persons to obtain a second psychological score.

[0078] Furthermore, the specific steps of constructing the depression detection and evaluation model in this embodiment include:

[0079] B1. Through data capture technology, we captured public depression detection videos and text data, and obtained the facial image data of the recovered patients shot by the remote guidance system. We then used the OPenCV tool to extract frames from the collected video data, performed filter enhancement preprocessing on the extracted images, and performed word segmentation preprocessing on the text data.

[0080] B2. Annotate the enhanced image data with key feature points and corresponding depression score. In this embodiment, the annotated key feature points include the eyes, mouth, nose, cheeks and other positions, which can reflect changes in facial expressions. The annotations here are annotated by professionals using clinical depression rating scores. Mild depression: The total score is usually between 5 and 9 points, indicating mild depression symptoms; moderate depression: The total score is between 10 and 14 points, indicating moderate depression symptoms; severe depression: The total score is ≥15 points, indicating severe depression symptoms;

[0081] B3. Constructing a depression detection and evaluation model, see Figure 2 , input the annotated image data into the image encoding layer of the evaluation model to obtain facial highlight image features, input the preprocessed text data into the text encoding layer to obtain text emotional features, and use the TCN encoding layer to extract time information features from the original video data; further, in this embodiment, the image encoding layer is constructed by a pre-trained Contrastive Language-Image Pre-Training sub-model, which is developed by OpenAI and aims to learn the common representation between images and texts through large-scale pre-training; the text encoding layer is constructed by a pre-trained EmoBERTa sub-model, where EmoBERTa is a pre-trained model based on the BERT architecture, which is specially optimized for understanding and generating text with emotional colors; the TCN encoding layer is constructed by a bidirectional TCN network, and the expansion factor d corresponding to the bidirectional TCN network is 8;

[0082] B4. Input the acquired facial highlight image features, text emotion features and time information features into the collaborative fusion layer constructed by the collaborative attention network to perform feature fusion and obtain depression fusion features;

[0083] B5. Input the obtained depression fusion features into the output layer of the evaluation model to obtain the depression evaluation level and the corresponding evaluation score; in this embodiment, the output layer is constructed by a two-layer fully connected network;

[0084] B6. Use the output evaluation score and the true labeled score to construct a cross entropy function, set the training threshold to 0.05, and set the training cycle to 100 to train the depression detection evaluation model to obtain the trained depression detection evaluation model;

[0085] B7. Configure the trained depression detection and evaluation model to the psychological evaluation unit, conduct professional depression detection and evaluation on the recovered person, and obtain the second psychological score;

[0086] B7. Calculate the second health score of the recovered person based on the obtained first psychological score and the second psychological score.

[0087] Furthermore, in this embodiment, the physician portrait unit calculates the professional score and coachability level of the rehabilitation physician by calling the comprehensive fuzzy analysis algorithm configured in the physiological assessment unit. The specific steps include:

[0088] C1. Collect the professional level and work results data of rehabilitation physicians, and use the collected data on the professional level and work results of physicians to construct the first-level evaluation factor set and the second-level evaluation factor set. The first-level evaluation factor set includes education and qualifications, clinical experience, patient satisfaction, peer evaluation and scientific research results, among which: education and qualifications include the physician's educational background, professional qualification certificates, and scientific research results; clinical experience includes years of practice, the number and complexity of cases handled in a specific field (such as rehabilitation medicine); patient satisfaction includes patient feedback collected through questionnaires, online evaluations, etc.; scientific research results include published papers, research projects participated in, and scientific research awards received; the second-level evaluation factor set includes: the highest degree and professional certification status under education and qualifications; the success rate of specific disease treatment and complex case handling ability under clinical experience; satisfaction with treatment effects and communication skills evaluation under patient satisfaction; the number of high-quality papers and the total impact factor under scientific research results;

[0089] C2. Construct the evaluation levels that physicians can guide, including A, B, C, D, and E. Among them, A corresponds to the professional score of top rehabilitation physicians, corresponding to 80-100 points; B corresponds to the professional score of senior rehabilitation physicians, corresponding to 60-80 points; C corresponds to the professional score of intermediate rehabilitation physicians, corresponding to 40-60 points; D corresponds to the professional score of primary rehabilitation physicians, corresponding to 20-40 points; E corresponds to the professional score of achieved rehabilitation physicians, corresponding to 0-20 points;

[0090] C3. Based on the constructed first-level assessment factor set, second-level assessment factor set and physician coachability assessment level, repeat the A2-A6 process to calculate the physician professionalism score and coachability level.

[0091] Furthermore, the specific steps of the rehabilitation knowledge base unit in this embodiment include:

[0092] D1. Extract historical rehabilitation data from the historical rehabilitation case database, including basic patient information, diagnosis records, rehabilitation plans, treatment processes, treatment results and follow-up data, pre-process the extracted data to remove invalid, duplicate or erroneous information, and convert unstructured text information (such as medical record descriptions) into structured text data;

[0093] D2. Perform entity recognition and relationship extraction on the preprocessed text data through the configured pre-trained entity recognition and joint extraction algorithm, and construct recovery case triples using the extracted entities and relationships; in this embodiment, the entity recognition and joint extraction algorithm includes DeepKE tool, SciBERT or BioBERT;

[0094] D3, input the constructed rehabilitation case triples into the graph database to build a rehabilitation knowledge base;

[0095] D4, connecting the acquired historical rehabilitation training videos to the knowledge base triple data node corresponding to the historical rehabilitator through the interface of the constructed rehabilitation knowledge base and the corresponding historical rehabilitator name attribute;

[0096] D5. Extract the corresponding triple node data and relationship path data from the rehabilitation knowledge base, and input the extracted triple node data and relationship path data into the inference model for fine-tuning training to obtain a trained inference model; in this embodiment, the inference model includes existing models such as TransE, TransH, TransR and DistMult;

[0097] D6. Deploy the trained reasoning model to the rehabilitation knowledge base to obtain the rehabilitation reasoning knowledge base.

[0098] Furthermore, in this embodiment, the specific workflow steps of the rehabilitation matching planning unit include:

[0099] E1. Set a rehabilitation score selection threshold. When the obtained comprehensive rehabilitation score is lower than the rehabilitation score selection threshold, two selection buttons, namely rehabilitation physician and rehabilitation reasoning knowledge base, are provided through the guidance system interactive interface for the rehabilitation person to choose. In this embodiment, the rehabilitation score selection threshold corresponding to E1 is 13 points, and the rehabilitation guidance levels corresponding to the comprehensive rehabilitation score include excellent rehabilitation level, good rehabilitation level and medium rehabilitation level.

[0100] E2. When the patient selects a rehabilitation video generated by the rehabilitation reasoning knowledge base, the corresponding rehabilitation guidance video and rehabilitation guidance suggestions are recommended to the patient through a matching recommendation algorithm based on the obtained comprehensive rehabilitation score, the first health score, and the second health score of the patient;

[0101] E3. When the rehabilitation reasoning knowledge base does not contain rehabilitation videos and rehabilitation suggestions corresponding to the comprehensive rehabilitation score, the first health score and the second health score of the rehabilitated person, the reasoning model is called to generate corresponding rehabilitation guidance videos and rehabilitation guidance suggestions for the rehabilitated person through the existing rehabilitation knowledge base using the video generation algorithm and the text generation algorithm; in this embodiment, the video generation algorithm includes TensorFlow DCGAN for Video, Video Prediction using ConvLSTM, MoCoGAN and TGANv2; the text generation algorithm includes GPT, GPT-2, GPT-3, TensorFlow Text and PyTorchText;

[0102] E4. When the patient selects a rehabilitation physician, a matching threshold is set, and the matching coefficient between the patient and the rehabilitation physician is calculated through a matching recommendation algorithm based on the patient's comprehensive rehabilitation score, the first health score and the second health score, and the rehabilitation physician's professional score and coachability level. When the matching coefficient is greater than the matching threshold, the corresponding rehabilitation physician is recommended. If it is less than the matching coefficient, the matching continues until the matching is completed. If the rehabilitation physician whose matching coefficient is greater than the matching threshold cannot be matched after the matching is completed, the rehabilitation physician with the largest matching coefficient with the patient is recommended to the corresponding patient;

[0103] E5. Set a mental health threshold. When the second health score is greater than the mental health threshold, in addition to the rehabilitation physician recommended by E4, a corresponding mental health physician is recommended to the rehabilitated person across departments through a matching recommendation algorithm based on the second health score. If it is less than the mental health threshold, no mental health physician is recommended.

[0104] E6. When the obtained comprehensive rehabilitation score is higher than the rehabilitation score selection threshold, the matching recommendation algorithm is used to directly recommend the corresponding rehabilitation physician to the patient, and at the same time, the matching recommendation algorithm is used to recommend the mental health physician to the patient across departments according to the second health score; the rehabilitation guidance level corresponding to the comprehensive rehabilitation score E6 in this embodiment includes the level of rehabilitation that needs to be strengthened and the level of low rehabilitation;

[0105] E7. When the comprehensive rehabilitation score of the patient is lower than the rehabilitation score selection threshold after training guided by the rehabilitation physician and mental health physician recommended by E6, the patient will undergo the E1-E5 matching recommendation process.

[0106] This process significantly improves the allocation efficiency of rehabilitation resources and the treatment experience of rehabilitated patients through fine graded assessment and personalized recommendation technology. First, based on the rehabilitation score selection threshold, rehabilitated patients who need special intervention are screened out, and customized selection is provided through an intelligent interface, integrating the direct guidance of rehabilitation physicians and advanced rehabilitation knowledge base to ensure that each rehabilitated patient can obtain the most suitable rehabilitation plan for their condition; secondly, in the rehabilitation physician matching link, the matching coefficient and threshold setting are introduced to scientifically evaluate the fit between rehabilitated patients and physicians, ensuring that the recommended physicians are not only professional counterparts, but also can provide the highest quality guidance. For the concern about mental health status, the system sets an additional mental health threshold to timely identify and recommend psychological support resources, reflecting the concept of comprehensive care for the physical and mental health of rehabilitated patients. Finally, for rehabilitated patients with the greatest difficulty in rehabilitation, the system optimizes the recommendation strategy, not only matching the most suitable rehabilitation physician, but also integrating mental health resources across borders to ensure multi-dimensional support for the rehabilitation journey. This workflow not only strengthens the personalization and accuracy of rehabilitation services, but also effectively promotes the efficient allocation of rehabilitation resources and the overall improvement of rehabilitation quality through technological innovation and interdisciplinary cooperation.

[0107] Furthermore, the specific steps of the action monitoring unit scoring the standard of the training action include:

[0108] F1. Obtain rehabilitation physician video rehabilitation guidance data, standardized rehabilitation guidance rule text data and historical rehabilitation training video data through data capture technology, and perform denoising preprocessing on the collected data. At the same time, perform frame extraction on the preprocessed rehabilitation physician video rehabilitation guidance data and historical rehabilitation training video data through the OPenCV library to obtain rehabilitation training time series image data; in this embodiment, the standardized rehabilitation guidance rule text refers to the standard criteria for different limb rehabilitation movements formulated by professional rehabilitation physicians;

[0109] F2. Construct a training action evaluation model based on the Chinese pre-trained Bert sub-model and the pre-trained HRNet, and input the acquired rehabilitation training time series image data into the HRNet in the training action evaluation model to obtain the training action joint node height and curvature features. At the same time, input the pre-processed standardized rehabilitation guidance rule text data into the Chinese pre-trained Bert sub-model to extract the standard training rule threshold features, and input the acquired training action joint node height and curvature features and the acquired standard training rule threshold features into the cross-attention sub-model for fusion operation to obtain the fusion training standard features.

[0110] F3, input the acquired fusion training standard features into the output layer of the training action evaluation model, output the corresponding training action score, and use the extracted training action joint node height and curvature features and the standard training rule threshold features to calculate the training standard error, use the training standard error to construct a cross entropy function, train the training action evaluation model, and obtain the completed training action evaluation model;

[0111] F4. Deploy the training action evaluation model after training to the action monitoring unit. When the rehabilitated person is trained by the rehabilitation physician or rehabilitation video recommended by the guidance system, the deployed sensors collect the training actions of the rehabilitated person in real time, and upload the collected action data to the action monitoring unit for real-time scoring of the training action standard.

[0112] F5. Set a training action standardization score threshold. When the obtained real-time training action standardization score is less than the training action standardization score threshold, feedback is provided to adjust the training action in real time. Furthermore, in this embodiment, the training action standardization score threshold is set and obtained by expert experience method.

[0113] Furthermore, during training, some irregular movements may cause the training movements to be counterproductive, but the movement monitoring may sometimes be blocked and it is impossible to detect abnormal situations in the training process in time. The changes in human expressions are controlled by changes in human feelings. When training abnormalities occur, facial micro-expressions will also change accordingly. Therefore, in this embodiment, a micro-expression monitoring unit is used to monitor micro-expressions during training in real time to respond quickly to abnormal situations. The specific work flow of the micro-expression monitoring unit includes:

[0114] G1. Obtain an annotated public micro-expression database and historical training video data collected during rehabilitation training, use a pre-trained face detection model to extract and annotate the face area from the collected historical training video data, and integrate it with the acquired public micro-expression database; the pre-trained face detection model in this embodiment includes MTCNN and YOLOv5-face;

[0115] G2. The unified and integrated micro-expression data is subjected to key point labeling and image alignment, and the aligned image is subjected to filtering and enhancement processing to obtain a micro-expression training data set; key points include eyes, nose, corners of mouth, cheeks, etc.; the labels of this embodiment are positive micro-expressions, negative micro-expressions, and neutral micro-expressions; positive micro-expressions include:

[0116] Joy: A slight smile, fine lines around the eyes (crow's feet) appear, the upper lip is lifted, and the cheeks are bulged, showing a relaxed and happy mood;

[0117] Excitement: eyes slightly widened, eyebrows slightly raised, may be a quick reaction to good news or something of interest;

[0118] Anticipation: A slight forward lean with eyes slightly bright, showing curiosity or optimism about what is about to happen;

[0119] Satisfaction: The corners of the mouth are slightly raised and the brows are relaxed, showing satisfaction with a certain result or situation.

[0120] Negative micro-expressions include:

[0121] Negative microexpressions are associated with negative, uncomfortable, or unpleasant emotions and are often more pronounced when individuals try to hide or suppress their true feelings. They include:

[0122] Disgust: wrinkling of the nose, lifting of the upper lip, possibly with a slight tilt of the head back, to express dislike for a perception, taste, or idea;

[0123] Anger: Eyebrows drooped, eyes widened or narrowed, lips pursed or teeth bitten, showing dissatisfaction or anger;

[0124] Fear: eyes widen, pupils dilate, brows furrow, jaw may tighten slightly, reflecting fear or worry;

[0125] Sadness: The corners of the mouth droop, the brows wrinkle, and the eyes may droop slightly or have tears in them, indicating inner loss or sadness;

[0126] Pain: Eyebrows tend to pull down and draw inward, forming a "pain brow" where the brows are furrowed and deep wrinkles form between the brows. The cheeks may tense from clenching the teeth, the jaw line may appear more tense, and there may even be a tremor.

[0127] G3, construct a micro-expression detection model, input the images in the acquired micro-expression training data set into the initial feature extraction layer in the micro-expression detection model in order of time points, and obtain the key features of facial expressions; in this embodiment, the initial feature extraction layer includes ResNet, MobileNet, and FaceNet;

[0128] G4. Input the key features of the acquired facial expression into the optical flow sub-model layer in the micro-expression detection model to obtain the dynamic change features of the local micro-expressions in the key features of the facial expression; the dynamic change features of the local micro-expressions include the rising and falling of the corners of the mouth, the twitching of the nose, the movement of the cheeks, the rising and falling of the eyebrows, and other changing movements; the optical flow sub-model layer in this embodiment is constructed by the existing streamer model, which is used to describe the motion information of objects or surfaces in space in the image sequence;

[0129] G5, input the key features of facial expressions and the dynamic change features of local micro-expressions into the collaborative attention layer, perform feature fusion, and obtain fused micro-expression features;

[0130] G6. Input the acquired fused micro-expression features into the classification layer in the model, output the probability of detecting the corresponding type of micro-expression, set the cross entropy function as the model loss function, set the training cycle to 200, train the micro-expression detection model, and obtain the trained micro-expression detection model; in this embodiment, the classification layer is constructed by the softmax function;

[0131] G7. Deploy the acquired micro-expression detection model to the micro-expression monitoring unit. When the rehabilitated person is training, the sensor collects the facial image or video data of the rehabilitated person in real time, transmits the collected data to the micro-expression monitoring unit for detection, and outputs the micro-expression detection results.

[0132] G8. Set an abnormal probability threshold. When the probability that the detected output micro-expression is a negative micro-expression is greater than the abnormal probability threshold, an alarm feedback is performed, otherwise no alarm feedback is performed; in this embodiment, the abnormal probability threshold is 0.8.

[0133] Example 2

[0134] See also Figure 3 Another embodiment provided by the present invention: A specific workflow of a module corresponding unit in an intelligent remote rehabilitation guidance system includes:

[0135] First, the rehabilitated person and the rehabilitation physician log in and upload corresponding information through the human-computer interaction module, and encrypt the uploaded information through the encryption module; secondly, the first health data of the rehabilitated person is collected through the physiological evaluation unit, and the first health score is calculated using the comprehensive fuzzy evaluation algorithm configured in the unit; at the same time, the psychological data of the rehabilitated person is collected and analyzed through the psychological evaluation unit, and the second health score is obtained using the emotion and depression detection algorithm configured in the unit; thirdly, the rehabilitation level portrait unit calculates the comprehensive rehabilitation score of the rehabilitated person according to the first health score and the second health score, and divides the rehabilitation guidance level accordingly; further, in this embodiment, the first health score is a physiological health score, and the second health score is a psychological health score;

[0136] Fourth, the physician data is obtained through the physician portrait unit, and the professional score of the rehabilitation physician is evaluated through the configured comprehensive fuzzy evaluation algorithm. The rehabilitation guidance level that the rehabilitation physician can provide is determined based on the professional score. At the same time, the rehabilitation knowledge base unit builds a rehabilitation knowledge base through visual algorithms and natural language algorithms based on the historical rehabilitation case data obtained, and obtains the rehabilitation reasoning knowledge base based on the built-in reasoning algorithm; Fifth, the rehabilitation matching planning unit recommends matching rehabilitation physicians or health planning guidance opinions generated by the rehabilitation reasoning knowledge base to the rehabilitation patient based on the rehabilitation patient's comprehensive rehabilitation score, economic status, physician portrait and rehabilitation reasoning knowledge base through a matching recommendation algorithm;

[0137] Sixth, during the rehabilitation process, the standard of the rehabilitation training movement is monitored in real time through the action monitoring unit, and the standard of the training movement is scored, and the score is fed back to the intelligent interaction unit. At the same time, the micro-expression monitoring unit is used to monitor the changes of the micro-expressions of the rehabilitation during the training process in real time. When the monitored micro-expressions change abnormally, the monitoring results are fed back to the intelligent interaction unit. The intelligent interaction unit gives the rehabilitation person corresponding feedback, encouragement and adjustment suggestions based on the training movement standard score and micro-expression changes monitored during the rehabilitation training process by configuring the intelligent rehabilitation assistant, and feeds back the non-standard training movement or training abnormality that occurs during the training process to the corresponding recommended rehabilitation physician, and adjusts or stops the training movement and training process in real time, so as to improve the rehabilitation person's participation and the scientific nature of the training process;

[0138] Seventh, the rehabilitation assessment unit collects the rehabilitation training data of each patient and the physiological and psychological data after the training, uses the machine learning algorithm to predict the change trajectory of the patient's comprehensive rehabilitation score, and feeds the collected comprehensive rehabilitation score change trajectory back to the rehabilitation matching planning unit through the rehabilitation advice and feedback unit to adjust the subsequent rehabilitation plan, ensure that the rehabilitation process is more personalized and accurate, and promote the patient's continuous rehabilitation.

[0139] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation modes, which are merely illustrative rather than restrictive. Under the guidance of the present invention, ordinary technicians in the field may also change, modify, replace and modify the above-mentioned embodiments without departing from the scope of protection of the purpose of the present invention and the claims, and all of these are within the protection of the present invention.

Claims

1. An intelligent remote rehabilitation guidance system, characterized in that: include: Rehabilitation patient portrait module, rehabilitation planning module, and remote interactive monitoring module; The rehabilitation patient portrait module includes a physiological assessment unit, a psychological assessment unit and a rehabilitation level portrait unit; The physiological evaluation unit is used to obtain the first health data of the rehabilitated person and calculate the first health score of the rehabilitated person through the configured comprehensive fuzzy evaluation algorithm; The psychological assessment unit is used to obtain the second rehabilitation data of the rehabilitated person, and detect and calculate the second health score of the rehabilitated person through the configured psychological questionnaire and depression detection algorithm; the rehabilitation level portrait unit is used to calculate the comprehensive rehabilitation score of the rehabilitated person according to the calculated first health score and second health score, and classify the rehabilitation guidance level of the rehabilitated person according to the obtained comprehensive rehabilitation score; The rehabilitation planning module includes a physician portrait unit, a rehabilitation knowledge base unit and a rehabilitation matching planning unit; The physician portrait unit is used to obtain physician professional level and work achievement data, call the comprehensive fuzzy evaluation algorithm in the physiological evaluation unit, and calculate the professional score and coachability level of the rehabilitation physician; the rehabilitation knowledge base unit is used to obtain historical rehabilitation case library data, build a rehabilitation knowledge base through the configured natural language algorithm, and build a reasoning model into the constructed rehabilitation knowledge base to obtain a rehabilitation reasoning knowledge base with reasoning ability; The rehabilitation matching planning unit is used to recommend a matching rehabilitation physician across departments to the rehabilitated person according to the obtained comprehensive rehabilitation score of the rehabilitated person through a matching recommendation algorithm, or to recommend a matching rehabilitation knowledge base to the rehabilitated person through a matching recommendation algorithm and an online health guidance opinion or guidance video generated by an intelligent generation algorithm; The remote interactive monitoring module includes a motion monitoring unit and a micro-expression monitoring unit; the motion monitoring unit is used to monitor the standard of the training motion of the rehabilitator in real time according to the configured training motion evaluation algorithm, and to provide scoring feedback on the standard of the training motion; The micro-expression monitoring unit is used to monitor the changes in micro-expressions of the rehabilitator during training in real time according to the configured micro-expression monitoring algorithm, and to provide feedback on the monitoring results when the monitored micro-expressions undergo abnormal changes.

2. The intelligent remote rehabilitation guidance system according to claim 1, characterized in that: The specific steps of the physiological assessment unit calculating the first health score of the recovered person include: A1. Use various physiological monitoring devices to collect the real-time first health data of the rehabilitated person, clean the collected data, remove abnormal values, standardize the preprocessing, and obtain the first health data after preprocessing; A2. Constructing a first-level evaluation factor set s=[s1…s l …s L ], and use the sub-factors corresponding to the first health factor set to construct the secondary evaluation factor set s l =[s l1 …s lk …s lK ],s l represents the lth first-level evaluation factor in the first-level evaluation factor set, s lk represents the kth secondary evaluation factor in the lth primary evaluation factor; L represents the length of the primary evaluation factor set, and K represents the length of the secondary evaluation factor set corresponding to the lth primary evaluation factor; A3. Set the rehabilitation index level set to v = [v1…v m …v M ], evaluate and score the constructed first-level evaluation factor set and second-level evaluation factor set by expert scoring, and obtain the fuzzy evaluation matrix R of the first-level evaluation factor set and the fuzzy evaluation matrix R' of the second-level evaluation factor set in turn; where v m represents the mth rehabilitation index level, and M represents the total number of rehabilitation index levels.

3. An intelligent remote rehabilitation guidance system as claimed in claim 2, characterized in that: The specific steps of the physiological assessment unit calculating the first health score of the recovered person also include: A4. According to the obtained R', calculate the weight W of the second-level evaluation factor under the lth first-level evaluation factor through the minimum membership weighted average deviation l =[w l1 …w lk …w lK ], and calculate the comprehensive evaluation vector C of K secondary evaluation factors l =[C l1 …C lm …C lM ], as follows: where w lk represents the weight of the kth secondary evaluation factor under the lth primary evaluation factor, c lm represents the comprehensive evaluation value of the second-level evaluation factor under the lth first-level evaluation factor at the mth rehabilitation index level, g k represents the maximum membership of the kth secondary assessment factor and the rehabilitation indicator level set; r' km It indicates the membership degree of the k-th secondary assessment factor to the m-th rehabilitation indicator level; A5. Based on the secondary assessment factors l The comprehensive evaluation vector C of the evaluation index level l , calculate the first-level evaluation factor index weight W s , by obtaining W s =(w1…w l …w L ), using the fuzzy evaluation matrix R, W of the first-level evaluation factor set s The comprehensive evaluation vector of the first-level evaluation factor is calculated by the comprehensive evaluation result formula of a single first-level evaluation factor A6. Comprehensively evaluate the calculated first-level evaluation factors Add up all the elements to get the first health score of the recovered person.

4. The intelligent remote rehabilitation guidance system according to claim 3, characterized in that: The specific workflow steps of the rehabilitation matching planning unit include: E1. Set a rehabilitation score selection threshold. When the comprehensive rehabilitation score obtained is lower than the rehabilitation score selection threshold, two selection buttons, namely rehabilitation physician and rehabilitation reasoning knowledge base, are provided through the guidance system interactive interface for the rehabilitation patient to choose. E2. When the patient selects a rehabilitation video generated by the rehabilitation reasoning knowledge base, the corresponding rehabilitation guidance video and rehabilitation guidance suggestions are recommended to the patient through a matching recommendation algorithm based on the patient's comprehensive rehabilitation score, the first health score, and the second health score; E3. When the rehabilitation reasoning knowledge base does not contain rehabilitation videos and rehabilitation suggestions corresponding to the comprehensive rehabilitation score, first health score and second health score of the rehabilitated person, the reasoning model is called to generate corresponding rehabilitation guidance videos and rehabilitation guidance suggestions for the rehabilitated person through the existing rehabilitation knowledge base using the video generation algorithm and the text generation algorithm.

5. The intelligent remote rehabilitation guidance system according to claim 4, characterized in that: The specific workflow steps of the rehabilitation matching planning unit also include: E4. When the patient selects a rehabilitation physician, a matching threshold is set, and the matching coefficient between the patient and the rehabilitation physician is calculated through a matching recommendation algorithm based on the patient's comprehensive rehabilitation score, the first health score and the second health score, and the rehabilitation physician's professional score and coachability level. When the matching coefficient is greater than the matching threshold, the corresponding rehabilitation physician is recommended. If it is less than the matching coefficient, the matching continues until the matching is completed. If the rehabilitation physician whose matching coefficient is greater than the matching threshold cannot be matched after the matching is completed, the rehabilitation physician with the largest matching coefficient with the patient is recommended to the corresponding patient; E5. Set a mental health threshold. When the second health score is greater than the mental health threshold, in addition to the rehabilitation physician recommended by E4, the corresponding mental health physician is recommended to the rehabilitated person across departments through a matching recommendation algorithm based on the second health score. If it is less than the threshold, no mental health physician is recommended.

6. The intelligent remote rehabilitation guidance system according to claim 5, characterized in that: The specific workflow steps of the rehabilitation matching planning unit also include: E6. When the comprehensive rehabilitation score obtained is higher than the rehabilitation score selection threshold, the matching recommendation algorithm is used to directly recommend the corresponding rehabilitation physician to the patient, and at the same time, the matching recommendation algorithm is used to recommend a mental health physician to the patient across departments based on the second health score; E7. When the comprehensive rehabilitation score of the patient is lower than the rehabilitation score selection threshold after training guided by the rehabilitation physician and mental health physician recommended by E6, the patient will undergo the E1-E5 matching recommendation process.

7. The intelligent remote rehabilitation guidance system according to claim 6, characterized in that: The specific steps of the action monitoring unit scoring the training action standard include: F1. Obtain rehabilitation physician video rehabilitation guidance data, standardized rehabilitation guidance rule text data and historical rehabilitation training video data, and perform denoising preprocessing on the collected data. At the same time, perform frame extraction on the preprocessed rehabilitation physician video rehabilitation guidance data and historical rehabilitation training video data through the OPenCV library to obtain rehabilitation training time series image data; F2. Construct a training action evaluation model based on the Chinese pre-trained Bert sub-model and the pre-trained HRNet, and input the acquired rehabilitation training time series image data into the HRNet in the training action evaluation model to obtain the training action joint node height and curvature features. At the same time, input the pre-processed standardized rehabilitation guidance rule text data into the Chinese pre-trained Bert sub-model to extract the standard training rule threshold features, and input the acquired training action joint node height and curvature features and the acquired standard training rule threshold features into the cross-attention sub-model for fusion operation to obtain the fusion training standard features. F3. Input the acquired fusion training standard features into the output layer of the training action evaluation model, output the corresponding training action score, and use the extracted training action joint node height and curvature features and the standard training rule threshold features to calculate the training standard error. Use the training standard error to construct a cross entropy function, train the training action evaluation model, and obtain the completed training action evaluation model.

8. The intelligent remote rehabilitation guidance system according to claim 7, characterized in that: The specific step of the action monitoring unit scoring the training action standard degree also includes: F4. Deploy the training action evaluation model after training to the action monitoring unit. When the rehabilitated person is trained by the rehabilitation physician or rehabilitation video recommended by the guidance system, the deployed sensors collect the training actions of the rehabilitated person in real time, and upload the collected action data to the action monitoring unit for real-time scoring of the training action standard. F5. Set the training action standard score threshold. When the obtained real-time training action standard score is less than the training action standard score threshold, feedback is given and the training action is adjusted in real time.

9. An intelligent remote rehabilitation guidance system as claimed in claim 8, characterized in that: The specific workflow of the micro-expression monitoring unit includes: G1. Obtain an annotated public micro-expression database and historical training video data collected during rehabilitation training, use a pre-trained face detection model to extract and annotate the face area from the collected historical training video data, and integrate it with the obtained public micro-expression database; G2, perform key point labeling and image alignment on the unified and integrated micro-expression data, and perform filtering and enhancement processing on the aligned images to obtain a micro-expression training data set; G3, construct a micro-expression detection model, input the images in the acquired micro-expression training data set into the initial feature extraction layer of the micro-expression detection model in order of time points, and obtain the key features of facial expressions; G4, input the key features of the acquired facial expression into the optical flow sub-model layer in the micro-expression detection model to obtain the dynamic change features of the local micro-expressions in the key features of the facial expression; G5, input the key features of facial expressions and the dynamic change features of local micro-expressions into the collaborative attention layer, perform feature fusion, and obtain fused micro-expression features; G6. Input the acquired fused micro-expression features into the classification layer in the model, output the probability of detecting the corresponding type of micro-expression, set the cross entropy function as the model loss function, and set the training cycle to 200 to train the micro-expression detection model to obtain the trained micro-expression detection model.

10. The intelligent remote rehabilitation guidance system according to claim 9, characterized in that: The specific workflow of the micro-expression monitoring unit also includes: G7. Deploy the acquired micro-expression detection model to the micro-expression monitoring unit. When the rehabilitated person is training, the sensor collects the facial image or video data of the rehabilitated person in real time, transmits the collected data to the micro-expression monitoring unit for detection, and outputs the micro-expression detection results. G8. Set an abnormal probability threshold. When the probability that the detected output micro-expression is a negative micro-expression is greater than the abnormal probability threshold, an alarm feedback is performed, otherwise no alarm feedback is performed.

Images