An intelligent leg training and rehabilitation method

Abstract

The present invention discloses an intelligent leg training and rehabilitation method, including constructing a leg training and rehabilitation system based on a deep learning model and a data mining and analysis algorithm; analyzing the healing degree of the patient's rehabilitation training according to the output of the leg training and rehabilitation system, and customizing the leg rehabilitation training content for the patient; monitoring the performance and progress of the patient during the rehabilitation training through the leg training and rehabilitation system, and adjusting the leg rehabilitation training content; based on the results monitored by the leg training and rehabilitation system, introducing an expression analysis module to monitor the patient's expression state and perform human intervention; evaluating and optimizing the training content of the leg training and rehabilitation system; to avoid over-reliance on deep learning and data mining algorithms, introducing an expression analysis module to provide real-time expression state analysis, and adjusting the training and rehabilitation strategy in real time according to the analysis results, providing leg rehabilitation support more in line with the patient's expression state for the patient, and improving the rehabilitation speed of the patient's legs.

Description

Technical Field

[0001] The present invention belongs to the field of rehabilitation medicine in artificial intelligence, and specifically refers to an intelligent leg training and rehabilitation method. Background Art

[0002] The demand for rehabilitation medical services continues to rise. Although China's rehabilitation medical industry has established a three-level rehabilitation medical system, it still faces problems such as insufficient awareness and shortage of talents.

[0003] The development of artificial intelligence technology has brought new possibilities to rehabilitation medicine. AI technology can provide differentiated rehabilitation programs by analyzing a large amount of data, improving the efficiency and effectiveness of rehabilitation. For example, through deep learning algorithms, AI can predict the rehabilitation progress of patients and adjust the rehabilitation plan.

[0004] Chinese Patent Publication No.: CN1 17672457A discloses a micro-sensing feedback intelligent knee rehabilitation nursing method and device. The method includes: collecting the leg nerve signals and gait posture parameters of patients, filtering the leg nerve signals to obtain primary nerve signals; constructing a primary nerve signal feature extraction model to extract features from the primary nerve signals; using a rehabilitation gait correction model to generate knee care correction parameters and real-time correct the leg knee posture during the patient's walking process. The present invention performs multi-scale convolution features and spatial perception information fusion processing on the filtering results of leg nerve signals at different scales to obtain nerve signal features of spatial perception that are beneficial to characterizing nerve signal abnormalities, and generates correction parameters based on the gait posture parameters of patients according to the abnormality of the nerve signal feature distribution, realizing knee movement rehabilitation correction nursing based on the micro-sensing feedback of patients' leg nerves.

[0005] However, over-reliance on deep learning and data mining algorithms in rehabilitation medicine may lead to neglecting the importance of differentiated leg training and rehabilitation. Over-reliance on deep learning and algorithms easily ignores the facial expressions of patients' leg rehabilitation training and leg movement expressions, resulting in slow rehabilitation speed of patients. Summary of the Invention

[0006] In order to solve the problems in the above-mentioned prior art that over-reliance on deep learning and algorithms in rehabilitation medicine may lead to neglecting the importance of differentiated leg training and rehabilitation, over-reliance on deep learning and algorithms easily ignores the emotions and action expressions of patients, resulting in slow rehabilitation speed of patients, etc., the present invention proposes an intelligent leg training and rehabilitation method to improve the above problems.

[0007] Specifically, the present application is as follows:

[0008] An intelligent leg training and rehabilitation method includes the following steps:

[0009] S1: Based on a deep learning model and a data mining and analysis algorithm, construct a leg training and rehabilitation system, collect and analyze patient data;

[0010] S2: According to the output of the leg training and rehabilitation system, analyze the healing degree of the patient's rehabilitation training, create a rehabilitation training goal for each patient, and customize the leg rehabilitation training content in combination with the patient's leg rehabilitation training goal;

[0011] S3: Monitor the patient's performance and progress during the rehabilitation training through the leg training and rehabilitation system, and adjust the leg rehabilitation training content and goal according to the patient's performance and feedback;

[0012] S4: Based on the results monitored by the leg training and rehabilitation system, introduce an expression analysis module to monitor the patient's expression state and perform human intervention;

[0013] S5: Evaluate the leg rehabilitation training content of the leg training and rehabilitation system, and optimize the system based on the evaluation results.

[0014] Furthermore, the leg training and rehabilitation system at least includes a data analysis and prediction module, a leg rehabilitation training goal formulation module, a leg rehabilitation training monitoring and adjustment module, an expression analysis module, and an evaluation and optimization module;

[0015] The data analysis and prediction module is based on deep learning and data mining algorithms and is used to analyze patient data and predict the patient's performance and progress during leg rehabilitation training;

[0016] The leg rehabilitation training goal formulation module formulates a leg rehabilitation training goal and leg rehabilitation training content for each target patient according to the analysis results of all patient data in the system;

[0017] The leg rehabilitation training monitoring and adjustment module monitors the patient's performance and progress during leg rehabilitation training and adjusts the leg rehabilitation training content and goal;

[0018] The expression analysis module monitors the patient's expression state through the leg training and rehabilitation system;

[0019] The evaluation and optimization module is used to evaluate the leg rehabilitation training content of the leg training and rehabilitation system, and improve the system through the evaluation results, update patient data, and retrain and optimize the leg training and rehabilitation system.

[0020] Furthermore, the leg rehabilitation training goal formulation module specifically includes:

[0021] L1: Input the training feature data extracted from the historical leg rehabilitation training data of all patients in the system into the leg rehabilitation training analysis model, and use the data mining technology of the clustering algorithm and the cosine similarity algorithm to output the target recommended training content for the target patient;

[0022] L2: Determine the preference weights of the target patient for multiple training contents according to the historical training content data of the patient obtained from the system, generate a training content preference weight table, and calculate the recommendation values of multiple training rehabilitation courses in the target training rehabilitation course set according to the training content preference weight table and the attention value of each training rehabilitation course;

[0023] L3: Sort the recommendation values and recommend the course with the highest recommendation value to the patient;

[0024] The specific implementation of L2 is as follows:

[0025] Extract the first training duration of the training rehabilitation courses of the patient for each training content in the historical training content data, and extract the second training duration of the training rehabilitation courses of the patient for each training content within the target time range in the historical training content data. Calculate the preference weight of any training content of the patient through the following formula:

[0026] B = β1T1 + β2T2;

[0027] In the formula, B is the preference weight of the training content, T1 is the first training duration of the training content, T2 is the second training duration of the content format, and β1 and β2 are the first weight coefficient and the second weight coefficient respectively;

[0028] Calculate the preference weight of each training content through the above formula, generate a training content preference weight table, and calculate the recommendation value of any training rehabilitation course according to the following formula:

[0029] Q = BD;

[0030] In the formula, Q is the recommendation value of the training rehabilitation course, B is the preference weight of the training content to which the training rehabilitation course belongs, and D is the attention value of the training rehabilitation course. Calculate the recommendation value of each training rehabilitation course in the target training rehabilitation course set through the above formula;

[0031] The specific implementation of the leg rehabilitation training analysis model is as follows:

[0032] The leg rehabilitation training analysis model is obtained by training with a training dataset. For the historical training data of multiple patients in the leg training rehabilitation system, the time feature data of each patient in each training behavior is extracted, as well as the proportion of the training duration of the training rehabilitation course for each training content in the training behavior. The time feature data of each training behavior and the proportion of the training duration of the training rehabilitation course for each training content in the training behavior are associated and constructed to obtain a data subset, and a training set composed of multiple data subsets is generated. The leg rehabilitation training analysis model is trained through the training set to obtain a trained leg rehabilitation training analysis model.

[0033] Furthermore, in S2, the leg training rehabilitation system uses a data analysis algorithm based on the time series analysis method to analyze the healing degree of the patient's leg rehabilitation training. Among them, the specific analysis steps of the time series analysis method are as follows:

[0034] S21: Collect the healing degree data of the patient's leg rehabilitation training;

[0035] S22: Define the mean μ used to calculate the time series in the moving average;

[0036] S23: Use the moving average method to perform a moving average operation on the healing degree data of the patient's leg rehabilitation training;

[0037] S24: According to the analysis results, infer the change trend of the patient's leg rehabilitation training healing degree and formulate a leg rehabilitation training goal.

[0038] Furthermore, the specific calculation method of the moving average method is as follows:

[0039]

[0040] Among them, X t represents the moving average value at time point t; θ t-i+1 represents the data points of the healing degree of the leg rehabilitation training from time point t - i + 1 to t; q is the order of the model, μ is the mean of the time series, and γt is the random error term at time t.

[0041] Furthermore, in S3, the leg training rehabilitation system predicts the performance and progress of the patient during the leg rehabilitation training based on a deep learning model and compares it with the actual leg rehabilitation training performance and progress to judge the leg rehabilitation training effect of the patient. Among them, the specific calculation method of the deep learning model prediction is as follows: In the long short-term memory network LSTM:

[0042] f t = tanh(F hx ·x t + F hh ·f t-1+a h );

[0043] where x t represents the input data at time point t; f t-1 represents the forget state at time point t - 1; F hx represents the weight matrix input to the forget state; F hh represents the weight matrix from the forget state to the forget state; a h represents the bias term of the forget state; tanh represents the hyperbolic tangent activation function;

[0044] At each time point t, based on the forget state f t the progress of the leg rehabilitation training at the next time step is predicted, and the prediction of the output layer is expressed as:

[0045]

[0046] where represents the predicted progress of the leg rehabilitation training at time point t + 1; Fyh represents the weight matrix from the forget state to the output layer; a y represents the bias term of the output layer; Sigmoid represents the activation function.

[0047] Furthermore, in S4, the expression analysis module is based on the natural language understanding system and is used to analyze the expressions of patients;

[0048] where the natural language understanding system at least includes a visual analysis module, a speech analysis module, and an expression analysis module;

[0049] The visual analysis module captures the facial expressions and leg walking postures of patients through a camera; the speech analysis module uses a microphone device to capture the speech information of patients.

[0050] Furthermore, the expression analysis module analyzes the expression state of patients based on the facial expressions, leg walking postures, and speech data of patients collected by the visual analysis module and the speech analysis module. Among them, the multi-modal deep leg training rehabilitation model is specifically:

[0051] There are three inputs, namely facial expression data Q a , leg walking posture data Q b , and speech data Q c . Three branch neural networks are respectively defined. Let g(Q a ) represent the neural network output of the facial expression data branch, g(Q b ) represent the neural network output of the leg walking posture data branch, and g(Q c ) represent the neural network output of the speech data branch. Then:

[0052] D = Stack(g(Q a ), g(Q b ), g(Q c ));

[0053] P(Y = i|Q a , Q b , Q c ) = Sigmoid((m(D))i;

[0054] Where D represents a vector obtained by stacking the outputs of three branch neural networks g(Q a ), g(Q b ), g(Q c ) together on a new dimension; Stack represents stacking multiple tensors on a new dimension to generate a larger tensor; P(Y = i|Q a , Q b , Q c ) represents the probability that a sample belongs to the i-th class of expression given the facial expression data Q a , the leg walking posture data Q b and the speech data Q c ; m(D) represents a neural network layer that further processes the stacked tensor D; (m(D)) i represents the value of the i-th element in the vector output by the neural network model m; Sigmoid represents the activation function.

[0055] Furthermore, in the S5, the mean squared error loss function is used to measure the difference between the predicted value and the true value, and to evaluate the leg rehabilitation training content of the leg training rehabilitation system:

[0056]

[0057] Where Loss represents the loss function; T represents the number of steps in the time series data; N represents the total number of output types; x t+1，i represents the i-th element in the One-hot encoded form of the true leg rehabilitation training progress at time point t + 1; represents the predicted probability for the i-th category at time t + 1; w i is the weight for the i-th category.

[0058] Furthermore, the parameters of the model are adjusted by the Bayesian optimization algorithm to minimize the loss function and achieve the optimization and improvement of the system. The specific parameters adjusted by the Bayesian optimization algorithm are:

[0059] K m = K - a·δ KLoss;

[0060] Among them, K m represents the parameters of the updated model; a represents the leg training recovery rate, which is used to control the frequency of parameter update; δ K Loss represents the gradient of the loss function with respect to the parameter K; K represents the hyperparameter used to control the balance between model iteration and utilization.

[0061] The beneficial effects of an intelligent leg training and rehabilitation method of the present invention are as follows:

[0062] The present invention constructs a leg training and rehabilitation system based on a deep learning model and a data mining and analysis algorithm; analyzes the patient's rehabilitation training healing degree according to the output of the leg training and rehabilitation system, and customizes the patient's leg rehabilitation training content; monitors the patient's performance and progress during the rehabilitation training through the leg training and rehabilitation system, and adjusts the leg rehabilitation training content; based on the results monitored by the leg training and rehabilitation system, introduces an expression analysis module to monitor the patient's expression state and perform human intervention; evaluates and optimizes the training content of the leg training and rehabilitation system; to avoid over-reliance on deep learning and data mining algorithms, an expression analysis module is introduced to provide real-time expression state analysis, and the training and rehabilitation strategy is adjusted in real time according to the analysis results, providing leg rehabilitation support more in line with the patient's expression state for the patient, and improving the rehabilitation speed of the patient's legs. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 is a schematic flow chart of an intelligent leg training and rehabilitation method of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0064] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0065] Embodiment 1

[0066] Figure 1 is a schematic flow chart of an intelligent leg training and rehabilitation method of the present invention.

[0067] The method includes steps S1 - S5:

[0068] S1: Based on a deep learning model and a data mining and analysis algorithm, construct a leg training and rehabilitation system, and collect and analyze patient data;

[0069] Specifically, the patient data includes age, gender, historical medical record data, clinical registration data, clinical symptom report data, clinical diagnosis and test data, and clinical imaging data;

[0070] S2: Analyze the healing degree of the patient's rehabilitation training based on the output of the leg training and rehabilitation system, create rehabilitation training goals for each patient, and customize the leg rehabilitation training content in combination with the patient's leg rehabilitation training goals;

[0071] S3: Monitor the performance and progress of the patient during the rehabilitation training through the leg training and rehabilitation system, and adjust the leg rehabilitation training content and goals according to the patient's performance and feedback;

[0072] S4: Based on the results monitored by the leg training and rehabilitation system, introduce an expression analysis module to monitor the patient's expression state and perform human intervention;

[0073] S5: Evaluate the leg rehabilitation training content of the leg training and rehabilitation system, and optimize the system based on the evaluation results.

[0074] Furthermore, the leg training and rehabilitation system at least includes a data analysis and prediction module, a leg rehabilitation training goal formulation module, a leg rehabilitation training monitoring and adjustment module, an expression analysis module, and an evaluation and optimization module;

[0075] The data analysis and prediction module is based on deep learning and data mining algorithms, and is used to analyze patient data and predict the performance and progress of the patient during leg rehabilitation training;

[0076] The leg rehabilitation training goal formulation module formulates leg rehabilitation training goals and leg rehabilitation training content for each target patient according to the analysis results of all patient data in the system;

[0077] Specifically, both the leg rehabilitation training goals and the leg rehabilitation training content will be adjusted according to the patient's rehabilitation training progress, rehabilitation training healing degree, and needs;

[0078] The leg rehabilitation training monitoring and adjustment module monitors the performance and progress of the patient during leg rehabilitation training, and adjusts the leg rehabilitation training content and goals;

[0079] The expression analysis module monitors the patient's expression state through the leg training and rehabilitation system;

[0080] Specifically, the patient's expression state includes happy, joyous, excited, anxious, disgusted, or frustrated. If necessary, the system performs human intervention to promote the patient's emotional health and development, and ensure that the patient's emotional state is good during leg rehabilitation training;

[0081] The evaluation and optimization module is used to evaluate the leg rehabilitation training content of the leg training rehabilitation system, and improve the system based on the evaluation results, update the patient data, and retrain and optimize the leg training rehabilitation system.

[0082] Further, the leg rehabilitation training goal formulation module specifically includes:

[0083] L1: The training feature data extracted from the historical leg rehabilitation training data of all patients in the system is input into the leg rehabilitation training analysis model. Using data mining techniques of clustering algorithms and cosine similarity algorithms, the target recommended training content for the target patient is output;

[0084] Specifically, the clustering algorithm is used to classify the patient training features and match which type of features the target patient belongs to. Simply put, it is which type the specific condition of the target patient belongs to. Through big data, the recommended training content for the patient is further formulated quickly for this condition type. The clustering algorithm includes but is not limited to the K-Means algorithm or the DBSCAN algorithm.

[0085] L2: Determine the tendency weights of the target patient for multiple training contents based on the historical training content data of the patient obtained from the system, generate a training content tendency weight table, and calculate the recommendation values of multiple training rehabilitation courses in the target training rehabilitation course set according to the training content tendency weight table and the attention value of each training rehabilitation course;

[0086] L3: Sort the recommendation values and recommend the course with the highest recommendation value to the patient;

[0087] The specific implementation of L2 is as follows:

[0088] Extract the first training duration of the training rehabilitation courses of each training content of the patient in the historical training content data, and extract the second training duration of the training rehabilitation courses of each training content of the patient within the target time range in the historical training content data. Calculate the tendency weight of any training content of the patient through the following formula:

[0089] B = β1T1 + β2T2;

[0090] In the formula, B is the tendency weight of the training content, T1 is the first training duration of the training content, T2 is the second training duration of the content format, and β1 and β2 are the first weight coefficient and the second weight coefficient respectively;

[0091] Calculate the tendency weights of each training content through the above formula, generate a training content tendency weight table, and calculate the recommendation value of any training rehabilitation course according to the following formula:

[0092] Q = BD;

[0093] Wherein, Q is the recommended value of the training and rehabilitation course, B is the tendency weight of the training content to which the training and rehabilitation course belongs, and D is the attention value of the training and rehabilitation course. The recommended value of each training and rehabilitation course in the target training and rehabilitation course set is calculated through the above formula;

[0094] The specific implementation of the leg rehabilitation training analysis model is as follows:

[0095] The leg rehabilitation training analysis model is trained through a training data set. For the historical training data of multiple patients in the leg training and rehabilitation system, the time feature data of each patient in each training behavior and the proportion of the training duration of the training and rehabilitation course for each training content in the training behavior are extracted. The time feature data of each training behavior and the proportion of the training duration of the training and rehabilitation course for each training content in the training behavior are associated and constructed to obtain a data subset, and a training set composed of multiple data subsets is generated. The leg rehabilitation training analysis model is trained through the training set to obtain a trained leg rehabilitation training analysis model.

[0096] Further, in S2, the leg training and rehabilitation system uses a data analysis algorithm based on the time series analysis method to analyze the healing degree of the patient's leg rehabilitation training. The specific analysis steps of the time series analysis method are as follows:

[0097] S21: Collect the leg rehabilitation training healing degree data of the patient;

[0098] S22: Define the mean μ used to calculate the time series in the moving average;

[0099] S23: Use the moving average method to perform a moving average operation on the leg rehabilitation training healing degree data of the patient;

[0100] S24: According to the analysis result, infer the change trend of the patient's leg rehabilitation training healing degree and formulate a leg rehabilitation training goal.

[0101] Further, the specific calculation method of the moving average method is as follows:

[0102]

[0103] Wherein, X t represents the moving average value at time point t; θ t-i+1 represents the leg rehabilitation training healing degree data points from time point t - i + 1 to t; q is the order of the model, μ is the mean of the time series, and Yt is the random error term at time t.

[0104] Specifically, the moving average method is used to observe the overall trend of the patient's healing degree over time. Based on the results calculated in the moving average, trends such as the increase, decrease, or stability of the patient's healing degree can be observed, so as to better understand the patient's leg healing status, and combined with information such as the historical healing data, healing assessment data, and healing speed of other patients in the past, and then predict the patient's future leg rehabilitation training performance.

[0105] Further, in S3, the leg training and rehabilitation system predicts the performance and progress of the patient during the leg rehabilitation training based on a deep learning model, and compares it with the actual leg rehabilitation training performance and progress to judge the leg rehabilitation training effect of the patient. Among them, the specific calculation method predicted by the deep learning model is as follows: In the long short-term memory network LSTM:

[0106] f t = tanh(F hx ·x t + F hh ·f t-1 + a h );

[0107] Among them, x t represents the input data at time point t; f t-1 represents the forgetting state at time point t - 1; F hx represents the weight matrix input to the forgetting state; F hh represents the weight matrix from the forgetting state to the forgetting state; a h represents the bias term of the forgetting state; tanh represents the hyperbolic tangent activation function;

[0108] At each time point t, based on the forgetting state f t predict the leg rehabilitation training progress of the next time step, then the prediction of the output layer is expressed as:

[0109]

[0110] Among them, represents the predicted leg rehabilitation training progress at time point t + 1; F yh represents the weight matrix from the forgetting state to the output layer; a y represents the bias term of the output layer; Sigmoid represents the activation function.

[0111] In this embodiment, the leg training and rehabilitation system predicts the performance and progress of the patient during the rehabilitation training based on a deep learning model, and compares it with the actual rehabilitation training performance and progress to judge the rehabilitation training effect of the patient.

[0112] Further, in S4, the expression analysis module is based on a natural language understanding system and is used to analyze the expressions of patients;

[0113] Among them, the natural language understanding system at least includes a visual analysis module, a speech analysis module, and an expression analysis module;

[0114] The visual analysis module captures the facial expressions and leg walking postures of patients through a camera; the speech analysis module uses a microphone device to capture the speech information of patients.

[0115] Specifically, the speech information includes the speech rate, pitch, tone, and language expression features of patients; the expression analysis module can provide real-time expression state analysis, and adjust the patient's rehabilitation training strategy in real time according to the analysis results or provide support. Expression recognition helps to improve the interaction experience between patients and the artificial intelligence system, making the leg rehabilitation training process more user-friendly and meeting the needs of patients, improving the enthusiasm of patients. Through expression recognition, medical staff can more comprehensively evaluate the leg rehabilitation status and emotional state of patients, not only paying attention to the healing progress of patients, but also paying attention to the emotions and mental health status of patients. The expression state analysis at least includes emotional information.

[0116] In this embodiment, the natural language processing technology uses the mmPose-NLP method in extracting the leg walking posture features of patients. This is a sequence-to-sequence (Seq2Seq) skeleton key point estimator inspired by natural language processing. It uses millimeter wave (mmWave) radar data to accurately estimate up to 25 skeleton key points. This method voxelizes the radar point cloud (PCL) data, similar to the tokenization process in NLP. Then, the voxelized radar data frames are input into the mmPose-NLP architecture to predict the voxel indices of 25 bone key points, which are then converted back to real-world 3D coordinates. This process is similar to keyword extraction in NLP. Through keyword extraction analysis, it is determined that the current leg rehabilitation walking state of the patient is painful, comfortable, improving in rehabilitation, and not ideal in rehabilitation, and the patient's rehabilitation training strategy is adjusted in real time according to the analysis results or support is provided.

[0117] Further, the expression analysis module analyzes the expression state of patients based on the facial expressions, leg walking postures, and speech data of patients collected by the visual analysis module and the speech analysis module. Among them, the multi-modal deep leg training rehabilitation model is specifically:

[0118] There are three inputs, namely facial expression data Q a , leg walking posture data Q b , and speech data Q c . Three branch neural networks are respectively defined, and let g(Q aDenotes the neural network output of the facial expression data branch, g (Q b ) denotes the neural network output of the leg walking posture data branch, g(Q c ) denotes the neural network output of the speech data branch, then:

[0119] D = Stack(g(Q a ), g(Q b ), g(Q c ));

[0120] P(Y = i|Q a , Q b , Q c ) = Sigmoid((m(D))i;

[0121] Among them, D represents the vector obtained by stacking the outputs of the three branch neural networks g(Q a ), g(Q b ), g(Q c ) together on a new dimension; Stack represents stacking multiple tensors on a new dimension to generate a larger tensor; P(Y = i|Q a , Q b , Q c ) represents the probability that the sample belongs to the i-th class of expression given the facial expression data Q a , the leg walking posture data Q b and the speech data Q c ; m(D) represents the neural network layer that further processes the stacked tensor D; (m(D)) i represents the value of the i-th element in the vector output by the neural network model m; Sigmoid represents the activation function.

[0122] In this embodiment, the expression of the Sigmoid function is:

[0123]

[0124] Among them, Sigmoid(Q)i represents the output of the Sigmoid function acting on the i-th element of the vector Q, that is, represents the prediction probability of the i-th category; Q = [Q1, Q1,..., Q N represents a real-valued vector of length N, Q i represents the i-th element of the vector, usually corresponding to the linear transformation result of the neural network output layer; represents the exponential form of Qi; Denotes the exponential form summation of all elements in vector Q, which is used for the normalization process of the Sigmoid function; the Sigmoid function ensures that the sum of the output probabilities of all classes is 1. In multi-class classification problems, the output of the model can be interpreted as the predicted probability distribution of each class, and finally, the class with the highest probability is selected as the prediction result.

[0125] Furthermore, in S5, the mean squared error loss function is used to measure the difference between the predicted value and the true value, so as to evaluate the leg rehabilitation training content of the leg training rehabilitation system:

[0126]

[0127] where Loss represents the loss function; T represents the number of steps in the time series data; N represents the total number of output types; x t+1，i represents the i-th element in the One-Hot encoding form of the true leg rehabilitation training progress at time point t + 1; represents the predicted probability for the i-th class at time t + 1; w i is the weight of the i-th class.

[0128] Furthermore, the parameters of the model are adjusted by the Bayesian optimization algorithm to minimize the loss function, so as to optimize and improve the system. The specific parameters adjusted by the Bayesian optimization algorithm are:

[0129] Km = K - a·δ K LoSs;

[0130] where K m represents the parameters of the updated model; a represents the leg training rehabilitation rate, which is used to control the frequency of parameter update; δ K Loss represents the gradient of the loss function with respect to the parameter K; K represents the hyperparameter used to control the balance between model iteration and exploitation.

[0131] The present invention constructs a leg training rehabilitation system based on a deep learning model and a data mining analysis algorithm; analyzes the patient's rehabilitation training healing degree according to the output of the leg training rehabilitation system, and customizes the patient's leg rehabilitation training content; monitors the patient's performance and progress during the rehabilitation training through the leg training rehabilitation system, and adjusts the leg rehabilitation training content; based on the monitoring results of the leg training rehabilitation system, introduces an expression analysis module to monitor the patient's expression state and perform human intervention; evaluates and optimizes the training content of the leg training rehabilitation system; in order to avoid over-reliance on deep learning and data mining algorithms, an expression analysis module is introduced to provide real-time expression state analysis, and adjusts the training rehabilitation strategy in real time according to the analysis results, so as to provide the patient with leg rehabilitation support more in line with their expression state and improve the patient's leg rehabilitation speed.

[0132] The present invention and its implementation manners have been described above. Such description is not restrictive. What is shown in the drawings is only one of the implementation manners of the present invention, and the actual content is not limited thereto. In short, if those of ordinary skill in the art are inspired by it and, without departing from the gist of the present invention, creatively design structural manners and embodiments similar to the technical solution, they shall fall within the protection scope of the present invention.

Claims

1. An intelligent leg training rehabilitation method, characterized in that: The following steps are involved: S1: Based on deep learning models and data mining analysis algorithms, build a leg training rehabilitation system to collect and analyze patient data; The leg training rehabilitation system at least includes a data analysis and prediction module, a leg rehabilitation training target setting module, a leg rehabilitation training monitoring and adjustment module, an expression analysis module and an evaluation and optimization module; The data analysis and prediction module is based on deep learning and data mining algorithms to analyze patient data and predict the patient's performance and progress during leg rehabilitation training; S2: Analyze the patient's rehabilitation training healing degree according to the output of the leg training rehabilitation system, create rehabilitation training goals according to each patient, and customize the patient's leg rehabilitation training content in combination with the patient's leg rehabilitation training goals; The leg rehabilitation training target formulation module formulates leg rehabilitation training targets and leg rehabilitation training contents for each target patient according to the analysis results of all patient data in the system; S3: monitoring the patient's performance and progress during the rehabilitation training process through the leg training rehabilitation system, and adjusting the leg rehabilitation training content and goals according to the patient's performance and feedback; The leg rehabilitation training monitoring and adjustment module monitors the patient's performance and progress during the leg rehabilitation training and adjusts the content and goals of the leg rehabilitation training; S4: Based on the monitoring results of the leg training rehabilitation system, an expression analysis module is introduced to monitor the patient's expression state and perform human intervention; The expression analysis module monitors the patient's expression state through the leg training rehabilitation system; S5: Evaluate the leg rehabilitation training content of the leg training rehabilitation system, and optimize the system based on the evaluation result; The evaluation and optimization module is used to evaluate the leg rehabilitation training content of the leg training and rehabilitation system, and improve the system according to the evaluation results, update the patient data, and retrain and optimize the leg training and rehabilitation system; In S4, the expression analysis module is based on a natural language understanding system and is used to analyze the patient's expression; Wherein, the natural language understanding system includes at least a visual analysis module, a speech analysis module and an expression analysis module; The visual analysis module captures the patient's facial expression and leg walking posture through a camera; the voice analysis module uses a microphone device to capture the patient's voice information; The expression analysis module analyzes the patient's expression state based on the multimodal deep leg training rehabilitation model through the patient's facial expression, leg walking posture and voice data collected by the visual analysis module and the voice analysis module, wherein the multimodal deep leg training rehabilitation model is specifically: There are three inputs: facial expression data , Leg walking posture data and voice data , define three branch neural networks respectively, let g( ) represents the neural network output of the facial expression data branch, g( ) represents the neural network output of the leg walking posture data branch, g( ) represents the neural network output of the speech data branch, then: D=Stack(g( ),g( ),g( )); P(Y=i| , , )= ; Where D represents the three branch neural network g( ), g( ), g( )’s output is stacked together in a new dimension; Stack means that multiple tensors will be stacked in a new dimension to generate a larger tensor; P(Y=i| , , ) indicates that given facial expression data , Leg walking posture data and voice data In the case of , the probability that the sample belongs to the i-th type of expression; m(D) represents the neural network layer that further processes the stacked tensor D; Represents the value of the i-th element in the vector output by the neural network model m; Sigmoid represents the activation function.

2. An intelligent leg training rehabilitation method according to claim 1, characterized in that: The leg rehabilitation training target formulation module specifically includes: L1: The training feature data extracted from the historical leg rehabilitation training data of all patients in the system are input into the leg rehabilitation training analysis model, and the data mining technology of the clustering algorithm and the cosine similarity algorithm are used to output the target recommended training content for the target patient; L2: Determine the target patient's preference weights for multiple training contents based on the patient's historical training content data obtained from the system, generate a training content preference weight table, and calculate the recommended values ​​of multiple training and rehabilitation courses in the target training and rehabilitation course set based on the training content preference weight table and the attention value of each training and rehabilitation course; L3: Sort the recommendation values ​​and recommend the courses with the highest recommendation values ​​to patients; The specific implementation of L2 is: The first training duration of the patient's training rehabilitation course for each training content in the historical training content data is extracted, and the second training duration of the patient's training rehabilitation course for each training content within the target time range is extracted from the historical training content data, and the preference weight of any training content of the patient is calculated by the following formula: B = + ； In the formula, B is the inclination weight of the training content, is the first training duration of the training content, The second training duration for the content format, , are the first weight coefficient and the second weight coefficient respectively; The above formula is used to calculate the inclination weight of each training content, and a training content inclination weight table is generated. The recommended value of any training rehabilitation course is calculated according to the following formula: Q = BD; In the formula, Q is the recommended value of the training and rehabilitation course, B is the preference weight of the training content to which the training and rehabilitation course belongs, and D is the attention value of the training and rehabilitation course. The recommended value of each training and rehabilitation course in the target training and rehabilitation course set is calculated by the above formula; The leg rehabilitation training analysis model is specifically implemented as follows: The leg rehabilitation training analysis model is trained through the training data set. For the historical training data of multiple patients in the leg training rehabilitation system, the time characteristic data of each patient in each training behavior and the proportion of training time of the training rehabilitation courses for each training content in the training behavior are extracted. The time characteristic data of each training behavior and the proportion of training time of the training rehabilitation courses for each training content in the training behavior are associated to construct a data subset, and a training set composed of multiple data subsets is generated. The leg rehabilitation training analysis model is trained through the training set to obtain a trained leg rehabilitation training analysis model.

3. The intelligent leg training and rehabilitation method according to claim 1, characterized in that: In S2, the leg training rehabilitation system uses a data analysis algorithm based on a time series analysis method to analyze the healing degree of the patient's leg rehabilitation training, wherein the specific analysis steps of the time series analysis method are: S21: Collect the patient's leg rehabilitation training healing data; S22: Define the mean μ used to calculate the time series in the sliding average; S23: using a sliding average method to perform a sliding average operation on the patient's leg rehabilitation training healing degree data; S24: Based on the analysis results, infer the changing trend of the patient's leg rehabilitation training healing degree and set leg rehabilitation training goals.

4. The intelligent leg training and rehabilitation method according to claim 3, characterized in that: The specific calculation method of the sliding average method is: =μ+ + ; in, represents the sliding average at time point t; represents the leg rehabilitation training healing data points from time point t-i+1 to t; q is the order of the model, μ is the mean of the time series, is the random error term at time t.

5. The intelligent leg training and rehabilitation method according to claim 1, characterized in that: In S3, the leg training rehabilitation system predicts the patient's performance and progress in the leg rehabilitation training process based on the deep learning model, and compares it with the actual leg rehabilitation training performance and progress to determine the patient's leg rehabilitation training effect. The specific calculation method of the deep learning model prediction is: in the long short-term memory network LSTM: =tanh( · + · + ); in, represents the input data at time point t; represents the forgetting state at time point t-1; Represents the weight matrix input to the forget state; The weight matrix representing the forgotten state to the forgotten state; Represents the bias term of the forgetting state; tanh represents the hyperbolic tangent activation function; At each time point t, based on the forgetting state To predict the leg rehabilitation training progress at the next time step, the prediction of the output layer is expressed as: = Sigmoid ( · + ); in, represents the predicted leg rehabilitation training progress at time point t+1; Represents the weight matrix of the forgotten state to the output layer; Represents the bias term of the output layer; Sigmoid represents the activation function.

6. The intelligent leg training rehabilitation method according to claim 1, characterized in that: In S5, a mean square error loss function is used to measure the difference between the predicted value and the true value, so as to evaluate the leg rehabilitation training content of the leg training rehabilitation system: Loss = - ; Among them, Loss represents the loss function; T represents the number of steps in the time series data; N represents the total number of output types; The i-th element in the one-hot encoding form represents the actual leg rehabilitation training progress at time point t+1; represents the predicted probability for the i-th category at time t+1; is the weight of the ith category.

7. An intelligent leg training and rehabilitation method according to claim 6, characterized in that: The Bayesian optimization algorithm is used to adjust the parameters of the model to minimize the loss function and achieve optimization and improvement of the system. The specific adjustment parameters of the Bayesian optimization algorithm are: = K-a· Loss; in, represents the parameters of the updated model; a represents the leg training recovery rate, which is used to control the frequency of parameter updating; Loss represents the gradient of the loss function with respect to the parameter K; K represents a hyperparameter used to control the balance between model iteration and utilization.

Images