Severe rehabilitation training monitoring method and device

Abstract

The invention discloses a severe-illness rehabilitation training monitoring method and device, and relates to the technical field of severe-illness rehabilitation, and the method comprises the steps: collecting the motion data of a patient in real time, preliminarily analyzing the current motion state of the patient, carrying out the motion mode recognition of the patient, and outputting a preliminary motion mode label; inputting the motion mode labels into a generative adversarial network, training motion modes and motion postures in different stages through a generator, and generating virtual data with high similarity; a motion data set is initialized through virtual data, the motion data set is input into a multi-layer convolutional neural network architecture for feature extraction, and a motion rehabilitation state of the patient is generated; the training feedback data is uploaded to a cloud platform for storage and analysis, an early warning signal is generated according to a cloud data analysis result, and early warning is automatically given out when the training of the patient is abnormal. According to the invention, through a multi-level deep learning architecture and a virtual data generation method, the precision of motion pattern recognition and the evaluation effect of the rehabilitation state can be effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of critical illness rehabilitation, and in particular to a critical illness rehabilitation training monitoring method and device. Background Art

[0002] In recent years, with the development of rehabilitation medicine, especially in the exercise rehabilitation training of critically ill patients, real-time monitoring and evaluation of patients' movement status has gradually become a research hotspot. Traditional rehabilitation monitoring methods mostly rely on manual intervention or simple sensor equipment, which cannot accurately reflect the subtle posture deviations and changes in exercise load during the patient's exercise, making it difficult to maximize the rehabilitation effect. With the advancement of smart sensors, wearable devices and deep learning technology, more and more intelligent motion monitoring methods have been proposed and begun to be applied in the field of rehabilitation training. However, existing technologies generally have problems such as insufficient accuracy in motion pattern recognition, difficulty in motion data labeling and inaccurate real-time feedback, which restrict the improvement of rehabilitation training effects and the realization of personalized treatment plans.

[0003] Although deep learning-based motion monitoring methods have made some progress, they still face challenges such as missing data, inconsistent timing, and difficulty in capturing long-term dependencies between motion patterns when processing complex motion data. In addition, existing technologies make it difficult to accurately correct patients' motion postures and exercise loads at different stages, and lack sufficient attention to the real-time and accuracy of fatigue assessment. Therefore, how to efficiently integrate motion patterns, exercise loads, and physiological state data, and combine virtual data to enhance the diversity of training sets, has become the key to solving these problems. Summary of the invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides a critical care rehabilitation training monitoring method to solve the problem of how to accurately and in real time monitor the movement state of critically ill patients, correct the movement pattern, and dynamically adjust the exercise load in the training state.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions: In a first aspect, the present invention provides a critical care rehabilitation training monitoring method, which includes: collecting the patient's motion data in real time, preliminarily analyzing the patient's current motion state, performing motion pattern recognition on the patient, and outputting a preliminary motion pattern label; The motion pattern labels are input into the generative adversarial network, and the generator is used to train the motion patterns and motion postures at different stages to generate highly similar virtual data; Initialize the motion data set through virtual data, input the motion data set into a multi-level convolutional neural network architecture for feature extraction, and generate the patient's motor rehabilitation status; Based on the patient's motor rehabilitation status, the characteristics of different movement patterns are learned through deep neural networks. The multi-task learning framework is used to integrate the characteristics of different movement patterns into a unified training network. The feature layer in the shared model is used to establish the correlation between different tasks. Monitor the patient's training status between different tasks in real time, perform posture corrections based on changes in movement patterns, and assess the degree of fatigue during exercise to generate training feedback data; The training feedback data is uploaded to the cloud platform for storage and analysis, and a warning signal is generated based on the cloud data analysis results, and a warning is automatically issued when abnormalities occur in patient training.

[0007] As a preferred solution of the critical care rehabilitation training monitoring method of the present invention, wherein: the motion data includes motion posture data, motion load data and motion state evaluation data; The preliminary analysis of the patient's current motion state, identification of the patient's motion pattern, and output of a preliminary motion pattern label are specifically performed as follows: The motion posture data, motion load data and motion state assessment data are denoised through timestamp alignment and Kalman filtering; Use a temporal convolutional neural network to extract dynamic features from the processed motion posture data and load features from the processed motion load data; Using deep neural networks to extract rehabilitation features from processed movement state assessment data; The extracted dynamic features, load features, and rehabilitation features are input into the long short-term memory network to capture the temporal features and identify the patient's movement pattern; The patient's movement pattern is corrected in combination with real-time movement status assessment data, and a preliminary movement pattern label is output.

[0008] As a preferred solution of the critical care rehabilitation training monitoring method of the present invention, wherein: the movement pattern label is input into the generative adversarial network, and the movement patterns and movement postures of different stages are trained by the generator to generate highly similar virtual data, and the specific steps are as follows: The motion mode labels are used as input to the generator, where each label corresponds to a motion state in a time period; The generator uses a multi-layer structure that combines a bidirectional long short-term memory network with a self-attention mechanism to capture long-term and short-term dependencies in motion data. The different modal data in the motion data are mapped into a shared feature space through a cross-modal fusion layer, and the multi-layer structure of the generator is used to generate virtual data of motion modes at different stages. A deep convolutional neural network is used to build a discriminator to detect the authenticity of virtual data, and a temporal consistency loss is applied to ensure the temporal continuity of virtual data. Use virtual data generation models to supplement training datasets and pre-train the generator in combination with transfer learning methods; The motion pattern label is input into the pre-trained generator to generate virtual data, and after adversarial training of the discriminator, highly similar virtual data is output.

[0009] As a preferred solution of the critical care rehabilitation training monitoring method of the present invention, the specific steps of initializing the motion data set by virtual data, inputting the motion data set into a multi-level convolutional neural network architecture for feature extraction, and generating the patient's motion rehabilitation status are as follows: The generated virtual data is standardized and amplified to construct a motion data set, and is labeled based on the automatic label generation method of deep learning to form a motion training set; The motion training set is input into a multi-layer convolutional neural network, and the primary and high-level representations of dynamic features, load features, and rehabilitation features are gradually extracted through the convolutional layers; The extracted primary and high-level representations are input into the fully connected layer for fusion, and the patient's motor rehabilitation status is generated through the classification and regression network.

[0010] As a preferred solution of the critical care rehabilitation training monitoring method of the present invention, the following specific steps are used: based on the patient's motor rehabilitation status, the characteristics of different movement patterns are learned through a deep neural network, and the characteristics of different movement patterns are integrated into a unified training network using a multi-task learning framework. The feature layer in the shared model is used to establish the correlation between different tasks. According to the patient's motor rehabilitation status, the first task is to identify the change pattern of the movement posture, the second task is to evaluate the change trend of the exercise load, and the third task is to evaluate the abnormal state recognition of the patient's rehabilitation status; After clarifying the task objectives, a joint feature extraction layer is constructed by jointly building a convolutional neural network and a deep neural network. In the joint feature extraction layer, the features of different motion modes are extracted by the convolutional layer in the initial stage, and then further fused by the fully connected layer to form a joint feature space. In the joint feature space, the joint feature extraction layer adjusts the feature sharing ratio between each task in real time through the weighted attention mechanism, thereby establishing dynamic correlation between different tasks; Based on the data features of each task, the degree of association between each task and the joint feature extraction layer is adaptively adjusted. In the multi-task learning framework, a weighted loss function is used to adjust the loss weight of each task, and the contribution between tasks is optimized based on feedback during training. Combine the adjusted loss weights between tasks to establish dynamic correlation, construct the correlation matrix between tasks, and dynamically adjust the information flow between tasks accordingly; Based on the information flow between dynamically adjusted tasks, graph convolutional networks are used for modeling; According to the training process of each task, the weights of different tasks in the joint feature extraction layer are adjusted through real-time feedback. The real-time feedback in the training process is combined with the feedback signal generated by virtual data to further optimize the joint feature extraction layer of the task.

[0011] As a preferred solution of the critical care rehabilitation training monitoring method of the present invention, the real-time monitoring of the patient's training status between different tasks, performing posture correction for changes in movement patterns, and evaluating the degree of fatigue of the movement to form training feedback data are performed. The specific steps are as follows: Real-time monitoring of the patient's exercise posture, exercise load and physiological status data during training between different tasks, and denoising and time series alignment through Kalman filtering to output standardized training status data; Use temporal convolutional neural networks and long short-term memory networks to analyze standardized training state data and identify movement patterns and training task types in real time; Based on the movement mode and training task type, the posture deviation in the movement mode is detected and corrected in real time; Frequency domain analysis is used to extract heart rate, electromyography, respiratory rate and blood oxygen saturation from physiological state data as physiological signals; Input physiological signals into a deep neural network to evaluate fatigue status, and adjust the exercise load in the training state according to the fatigue evaluation; The adjusted training data is recorded in real time and integrated into training feedback data.

[0012] As a preferred solution of the critical care rehabilitation training monitoring method of the present invention, wherein: the training feedback data is uploaded to the cloud platform for storage and analysis, an early warning signal is generated according to the cloud data analysis results, and an early warning is automatically issued when abnormalities occur in the patient's training. The specific steps are as follows: Upload the training feedback data to the cloud platform through an encrypted channel for classified storage; Clean and denoise the classified and stored training feedback data, and extract the characteristics of exercise posture, exercise load and heart rate through deep learning analysis; The cloud platform identifies and marks abnormal patterns in training feedback data based on exercise posture, exercise load and heart rate characteristics. Based on the abnormal patterns, the cloud generates warning signals and divides the warning levels according to the severity; According to the warning level, notifications are issued through multiple channels for real-time warning.

[0013] In a second aspect, the present invention provides a critical care rehabilitation training monitoring device, including a data collection module, a pattern generation module, a feature extraction module, a feature integration module, a state monitoring module and a cloud warning module; The data collection module is used to collect the patient's motion data in real time, preliminarily analyze the patient's current motion state, identify the patient's motion pattern, and output a preliminary motion pattern label; The pattern generation module is used to input the motion pattern label into the generative adversarial network, and generate highly similar virtual data by training the motion patterns and motion postures at different stages through the generator; The feature extraction module is used to initialize the motion data set through virtual data, input the motion data set into a multi-level convolutional neural network architecture for feature extraction, and generate the patient's motor rehabilitation status; The feature integration module is used to learn the features of different movement patterns through a deep neural network based on the patient's motor rehabilitation status, integrate the features of different movement patterns into a unified training network using a multi-task learning framework, and establish the correlation between different tasks using the feature layer in the shared model; The state monitoring module is used to monitor the patient's training state between different tasks in real time, perform posture correction for changes in movement patterns, and evaluate the degree of fatigue during exercise to form training feedback data; The cloud-based early warning module is used to upload training feedback data to the cloud platform for storage and analysis, generate early warning signals based on cloud data analysis results, and automatically issue early warnings when abnormalities occur in patient training.

[0014] In a third aspect, the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, any step of the intensive care rehabilitation training monitoring method as described in the first aspect of the present invention is implemented.

[0015] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, any step of the critical care rehabilitation training monitoring method as described in the first aspect of the present invention is implemented.

[0016] The beneficial effects of the present invention are as follows: through a multi-level deep learning architecture and a virtual data generation method, the accuracy of motion pattern recognition and the evaluation effect of rehabilitation status can be effectively improved. First, by using a temporal convolutional neural network and a long short-term memory network to extract dynamic features and temporal features in motion data, it is possible to accurately capture subtle changes in motion posture, motion load, and rehabilitation status, thereby providing more accurate motion pattern labels. Secondly, by generating virtual data through a generative adversarial network and combining it with a multi-level convolutional neural network for training, it is possible to greatly enrich the training data set, improve generalization capabilities, and overcome the data scarcity problem that may occur during actual data collection. In addition, based on a multi-task learning framework, it is possible to dynamically adjust the weights between different motion pattern features, enhance the synergy between tasks, and further improve the correction accuracy of the motion pattern and the real-time nature of fatigue assessment. Finally, through the intelligent early warning function of the cloud platform, early warning signals can be issued in time when the patient's motion state is abnormal, which helps to achieve a personalized rehabilitation training program. BRIEF DESCRIPTION OF THE DRAWINGS

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

[0018] Figure 1 This is a flow chart of the critical care rehabilitation training monitoring method in Example 1.

[0019] Figure 2 This is a module diagram of the critical care rehabilitation training monitoring device in Example 1. DETAILED DESCRIPTION

[0020] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.

[0021] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0022] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0023] Example 1, reference Figure 1 and Figure 2 , which is the first embodiment of the present invention, and provides a critical care rehabilitation training monitoring method, comprising the following steps: S1: Collect the patient's motion data in real time, preliminarily analyze the patient's current motion status, identify the patient's motion pattern, and output preliminary motion pattern labels.

[0024] Furthermore, the motion data includes motion posture data, motion load data, and motion status assessment data; Specifically, the sensors worn on the patient (such as inertial measurement unit (IMU) sensors, force sensors, etc.) are used to collect motion posture data, motion load data, and motion state evaluation data in real time. Furthermore, the IMU sensor is used to collect the patient's three-dimensional acceleration, angular velocity, and magnetic field data to obtain the patient's motion posture information in real time; the motion load data is obtained in real time through the muscle electrical activity (EMG) sensor to evaluate the load changes of the patient during the training process; the motion state evaluation data is obtained in real time through physiological signal monitoring equipment (such as heart rate, respiratory rate, blood oxygen saturation, etc.) to evaluate the patient's physiological state.

[0025] Preliminary analysis of the patient's current motion state, identification of the patient's motion pattern, and output of preliminary motion pattern labels. The specific steps are as follows: The motion posture data, motion load data and motion status assessment data can be aligned with timestamps and de-noised using Kalman filtering to ensure high data quality and consistency, thereby accurately reflecting the patient's rehabilitation needs. Use a temporal convolutional neural network (TCN) to extract dynamic features from the processed motion posture data and load features from the processed motion load data; The use of temporal convolutional neural networks (TCNs) effectively solves the limitations of traditional neural networks in processing long time series data. Compared with traditional methods, TCNs have stronger temporal feature capture capabilities, can accurately identify small changes in patient movements, reduce movement deviations and errors, and thus reduce risks during training.

[0026] Using deep neural networks to extract rehabilitation features from processed movement state assessment data; The extracted dynamic features, load features, and rehabilitation features are input into the long short-term memory network (LSTM) to capture the temporal features and identify the patient's movement pattern; Specifically, the dynamic features, load features, and rehabilitation features are first arranged in a time series manner, and the corresponding time steps are constructed. Before being input into the LSTM network, the dynamic features, load features, and rehabilitation features are standardized to ensure that the dimensions of each feature are within the same order of magnitude to avoid unnecessary effects caused by excessive differences between features during network training. Then, the input of each time step is weighted through the LSTM gating mechanism to capture the temporal dependencies of the dynamic features, load features, and rehabilitation features, while retaining long-term memory information to avoid gradient vanishing or explosion problems. Finally, the results output by the LSTM network will identify the patient's movement pattern by analyzing the temporal features.

[0027] The patient's movement pattern is corrected in combination with real-time movement status assessment data, and a preliminary movement pattern label is output.

[0028] S2: Input the motion pattern label into the generative adversarial network, and use the generator to train the motion patterns and motion postures at different stages to generate virtual data with high similarity.

[0029] Furthermore, the motion mode labels are used as input to the generator, where each label corresponds to a motion state in a time period; Among them, these motion pattern labels carry key information related to the patient's motion pattern, such as the type of motion, load, posture, etc. Inputting these motion pattern labels into the generator provides the generator with guidance information about the motion state and helps the generator generate virtual data that matches it.

[0030] The generator uses a multi-layer structure that combines a bidirectional long short-term memory network with a self-attention mechanism to capture long-term and short-term dependencies in motion data. It should be noted that the motion pattern labels provide structural guidance for the motion data for the generator, while the complex network structure of the generator (a multi-layer structure of bidirectional long short-term memory network and self-attention mechanism) ensures the true simulation of the timing characteristics and motion state of the data. The two complement each other and jointly promote the generation of virtual data and the accurate reproduction of motion patterns.

[0031] The different modal data in the motion data are mapped into a shared feature space through a cross-modal fusion layer, and the multi-layer structure of the generator is used to generate virtual data of motion modes at different stages. Among them, the cross-modal fusion layer is a key component in deep learning for integrating data from different modalities; cross-modal fusion is widely used in multimodal learning tasks, especially in the fields of medical and motion monitoring. It can map multiple types of data into a shared feature space, thereby learning the potential associations between different data sources.

[0032] Furthermore, shared feature space refers to mapping different modal data in motion data (such as motion posture data, motion load data, and motion state evaluation data) into a unified representation space through a cross-modal fusion layer. This shared feature space is a high-dimensional feature space that contains comprehensive information from multiple data modalities, allowing data from different modalities to be correlated and compared with each other.

[0033] Preferably, the cross-modal fusion layer enhances the synergistic effect of multi-task learning by mapping motion data of different modalities (such as motion posture data, motion load data, and motion state assessment data) into a shared feature space, improves the processing capability and temporal consistency of generated data, and thus optimizes the accuracy and feedback effect of patient motor rehabilitation monitoring.

[0034] A deep convolutional neural network is used to build a discriminator to detect the authenticity of virtual data, and a temporal consistency loss is applied to ensure the temporal continuity of virtual data. It should be noted that temporal consistency loss is a loss function used in temporal data generation, which aims to ensure the smoothness and consistency of the generated data in the time dimension. Specifically, temporal consistency loss measures the continuity of the generated data between time steps and penalizes data that produce mutations or unnatural changes in the time series. It is often applied to tasks that process continuous time series, such as video generation, speech synthesis, and motion data modeling, to ensure that the data generated for each frame or time step remains coherent in time, rather than showing obvious inconsistencies or jumps.

[0035] Preferably, the temporal consistency loss calculates the changes between adjacent time steps of the generated data to ensure that there is a reasonable transition between each time point of the virtual data and the adjacent time points, avoiding abrupt fluctuations or time gaps in the generated data, and improving the naturalness and temporal consistency of the generated data, so that the virtual data generated during the training process is more consistent with the characteristics of real data in terms of long-term dependence and temporal continuity.

[0036] Use virtual data generation models to supplement training datasets and pre-train the generator in combination with transfer learning methods; Specifically, a preliminary training set is constructed by collecting and annotating a certain amount of real patient motion data, including multimodal data such as motion posture, motion load changes, and motion state evaluation. Then, these multimodal data are input into the deep learning model, and the temporal features and dynamic features of the motion data are extracted using the temporal convolutional neural network (TCN) and the long short-term memory network (LSTM). Next, based on these extracted temporal features and dynamic features, virtual data is generated through the generator of the generative adversarial network, but the initial generator may not be able to produce high-quality virtual data, so the transfer learning method is used.

[0037] Furthermore, the generator is pre-trained using existing large-scale motion datasets (such as public motion patterns or physiological signal datasets) to ensure that it can understand the basic laws and timing of motion patterns. During the pre-training process, the data patterns learned by the generator will be transferred to new tasks as prior knowledge, which can greatly shorten the training time and improve the quality of generated virtual data.

[0038] The motion pattern label is input into the pre-trained generator to generate virtual data, and then the discriminator is trained adversarially to output highly similar virtual data. Specifically, these pre-trained generators are used to virtually generate personalized patient data (i.e., generate virtual data), and the authenticity of the generated data is further optimized through adversarial training of the discriminator. At this point, the generator can generate motion data of multiple different stages based on the knowledge learned in pre-training, and ensure the temporal continuity of the virtual data through temporal consistency loss. Finally, these virtual data are combined with actual data to form a richer training data set (i.e., highly similar virtual data) for subsequent training of motion pattern recognition and rehabilitation status assessment models.

[0039] S3: Initialize the motion dataset through virtual data, input the motion dataset into a multi-level convolutional neural network architecture for feature extraction, and generate the patient's motor rehabilitation status.

[0040] Furthermore, the generated virtual data is standardized and amplified to construct a motion data set, and is labeled based on an automatic label generation method based on deep learning (such as a deep learning method based on a convolutional neural network (CNN) and a long short-term memory (LSTM) network) to form a motion training set; Specifically, the standardization process includes normalizing and scaling the data so that different types of motion data (such as motion posture data, motion load data, and motion status assessment data) can be processed within a unified range to ensure that they meet the input requirements for subsequent training.

[0041] Furthermore, data augmentation technology is used to further improve the diversity of the data set. This process includes deformation, translation, rotation and other operations on virtual data to simulate different motion conditions and scenarios (such as posture changes, load increase, training intensity changes, etc.), thereby generating more diverse and representative training samples to ensure that the data set covers all possible situations in actual training.

[0042] The motion training set is input into a multi-layer convolutional neural network, and the primary and high-level representations of dynamic features, load features, and rehabilitation features are gradually extracted through the convolutional layers; Among them, primary representations are usually low-level, specific pattern information, such as simple shapes of movement postures, basic fluctuations in load, etc., while high-level representations are more complex abstract patterns, such as the overall trend of movement status, changes in rehabilitation progress, etc. These primary and high-level representations capture various levels of information in patient movement data, providing a rich semantic basis for subsequent analysis.

[0043] The extracted primary and high-level representations are input into the fully connected layer for fusion, and the patient's motor rehabilitation status is generated through the classification and regression network.

[0044] Preferably, by fusing the extracted primary representation and high-level representation into the fully connected layer, it is possible to effectively combine low-level and high-level feature information, and enhance the comprehensive understanding and analysis of the patient's motor rehabilitation status. The fusion operation of the fully connected layer not only improves the information interaction between different feature levels, but also optimizes the recognition ability of complex movement patterns, thereby achieving more accurate rehabilitation status assessment.

[0045] Furthermore, by combining classification and regression networks, it is possible to simultaneously classify the patient's movement status (such as identifying different stages of rehabilitation) and regress continuous values ​​(such as assessing rehabilitation progress or fatigue level). This multi-task learning method that combines classification and regression can not only achieve more accurate movement pattern recognition, but also effectively predict the patient's rehabilitation effect and potential risks, providing more personalized and dynamic training feedback.

[0046] S4: Based on the patient's motor rehabilitation status, the characteristics of different movement patterns are learned through deep neural networks. The multi-task learning framework is used to integrate the characteristics of different movement patterns into a unified training network. The feature layer in the shared model is used to establish the correlation between different tasks.

[0047] Furthermore, according to the patient's motor rehabilitation status, task one is established to identify the changing pattern of the motor posture, task two is to evaluate the changing trend of the motor load, and task three is to evaluate the abnormal state identification of the patient's rehabilitation status; Preferably, the setting of different tasks reflects the multi-dimensional needs of patients in the rehabilitation process and ensures the comprehensiveness and accuracy of rehabilitation monitoring.

[0048] After clarifying the task objectives, a joint feature extraction layer is constructed by jointly building a convolutional neural network and a deep neural network. In the joint feature extraction layer, the features of different motion modes are extracted by the convolutional layer in the initial stage, and then further fused by the fully connected layer to form a joint feature space. Specifically, the convolution layer is used to extract local features from the input data, and the convolution kernel is usually used to slide the input data. Suppose the input data is , the output of the convolutional layer is , the convolution operation is specifically expressed by the following formula: ; in, Represents the feature map of the convolutional layer output, represents the convolution kernel, represents the bias term; Furthermore, the fully connected layer flattens the feature map output by the convolutional layer into a one-dimensional vector, multiplies it with the weight matrix, and adds the bias term, which is specifically expressed by the following formula: ; in, Represents a fully connected layer The weight matrix of Represents the feature map output by the convolutional layer Flattened into a one-dimensional vector, Represents a fully connected layer The bias term, represents the fusion features output by the fully connected layer, represents a nonlinear activation function; All fused features extracted and fused through convolution and fully connected layers will eventually be integrated into the joint feature space, which contains all features from different tasks.

[0049] Among them, the characteristics of the movement pattern include the characteristics of various movement behaviors and physiological reactions exhibited by the patient during the training process, such as movement posture, exercise load, movement amplitude, movement frequency, and physiological state during exercise (such as fatigue level, heart rate changes, etc.). These movement pattern characteristics can reflect the patient's movement method, training intensity, and rehabilitation progress during rehabilitation training. Through deep neural network analysis, it is possible to identify changes in different movement patterns and further guide the adjustment of exercise load and optimization of training plans.

[0050] It should be noted that the joint feature extraction layer refers to the extraction of features of input data by different tasks through a shared neural network layer in a multi-task learning framework. These joint feature extraction layers are usually located in the front part of the network and serve multiple different tasks. For example, tasks such as motion posture recognition, load assessment, and rehabilitation status recognition can share this part of the network layer to extract low-level and high-level features of the data in the same feature space. Through the joint feature extraction layer, the common features in the data can be better captured, the synergy between different tasks can be improved, and the learning of redundant features generated when each task is trained separately can be avoided.

[0051] In the joint feature space, the joint feature extraction layer adjusts the feature sharing ratio between each task in real time through the weighted attention mechanism, thereby establishing dynamic correlation between different tasks; Dynamic relevance means that the dependencies between tasks in the shared feature layer will be automatically adjusted according to the training process and the needs of different tasks. For example, task A may be more sensitive to certain features in the shared feature space at a certain stage, while task B may be more dependent on these features at other stages.

[0052] Specifically, through the joint feature extraction of convolutional neural networks (CNN) and deep neural networks (DNN), preliminary features of multiple tasks are extracted and fused in the shared feature space. Then, the weighted attention mechanism dynamically adjusts the weight of each task in the shared feature layer based on the relevance of the current training state of each task. The feature sharing ratio of each task is not fixed, but changes according to the importance of tasks, training effects and relative needs of tasks during training. Furthermore, through this dynamic adjustment mechanism, the feature allocation between tasks can be optimized according to real-time feedback to ensure that each task obtains the optimal feature information in the shared feature space. Ultimately, this adjustment mechanism makes the relationship between tasks no longer static, but flexibly adjusted as the data characteristics and task objectives change during training, thereby establishing dynamic correlation between tasks and optimizing the synergy and overall performance between tasks.

[0053] The introduction of the weighted attention mechanism solves the problem of uneven feature sharing ratio between tasks in multi-task learning. By adjusting the feature sharing ratio between tasks in real time, resources can be dynamically allocated according to the importance of the task and the current training status, thereby improving learning efficiency and ensuring that the optimization goals of each task do not interfere with each other.

[0054] Based on the data features of each task, the degree of association between each task and the joint feature extraction layer is adaptively adjusted. In the multi-task learning framework, a weighted loss function is used to adjust the loss weight of each task and optimize the loss contribution between tasks. It should be noted that dynamic relevance adjusts the dependencies between tasks at the feature level, while adaptive adjustment adjusts the attention of each task to the shared feature layer based on the feedback of training data. The former uses a weighted attention mechanism to determine which features are more important for which tasks during training, while the latter further fine-tunes the feature extraction strategy of each task through changes in task features. Specifically, dynamic relevance provides a basic framework for subsequent adjustments, while adaptive adjustment further optimizes the framework based on the feedback of each task.

[0055] Specifically, firstly, based on the training data characteristics of each task, the different feature information required by each task is analyzed in order to understand the requirements of each task in the joint feature extraction layer. Then, by calculating the correlation between the data features of each task and the feature space output by the joint feature extraction layer, the importance of the task features in the shared feature space is evaluated. On this basis, the weighted attention mechanism is used to adaptively adjust the degree of association between each task and the joint feature extraction layer, so that the features of each task can be dynamically allocated more or less resources according to their importance. This adjustment is completed through the back-propagation mechanism, in which the loss function of each task is weighted according to its weight in the feature extraction layer, so that during the training process, the feature contribution of each task is optimized, and finally the feature sharing and synergy between tasks are maximized.

[0056] Furthermore, according to the feature importance and training progress of each task, the loss value of each task is calculated, and the loss of each task is weighted. The weighting coefficient is determined based on the priority of the task, the performance of the current training stage, and the relative importance of the tasks. Then, the weighted loss function is used to adjust the loss value of each task according to its corresponding weight, so that the contribution of the task in the training process is reasonably distributed. During the training process, as the loss of each task changes dynamically, the loss weight will be adaptively adjusted, and the weight distribution between tasks will be optimized through back propagation. Finally, the optimized loss weight not only improves the training effect of each task, but also ensures the balance of loss contribution between tasks, thereby enhancing the overall synergy of the multi-task learning framework.

[0057] For example, in a critical rehabilitation monitoring task, Task A (motion posture recognition) may be more dependent on fine motion features, while Task B (load assessment) needs to pay attention to the load change trend. By calculating the feature importance of each task, the weighted attention mechanism is used to adaptively adjust the degree of correlation between Task A and Task B in the shared feature space, thereby ensuring that Task A pays more attention to motion features and Task B pays more attention to load features. On this basis, in the multi-task learning framework, a weighted loss function is used to adjust the loss weight of each task. In the early stages of training, Task A may face higher losses, so it is given a higher weight. As the training progresses, the performance of Task B improves, and its loss weight is adjusted to maintain a balance between tasks. In this way, the loss contribution between tasks is optimized to ensure that each task receives appropriate attention and optimization at different stages during model training, ultimately improving overall performance.

[0058] Combine the adjusted loss weights between tasks to establish dynamic correlation, construct the correlation matrix between tasks, and dynamically adjust the information flow between tasks accordingly; Specifically, during the training process, the weights between tasks are optimized through real-time feedback, and the degree of dependence of each task on shared features is adjusted to ensure that the contribution between tasks meets the current training requirements. At this time, the low-level features of each task are extracted through the convolutional layer, and the self-attention mechanism is used to capture the similarities and dependencies between tasks, and the dynamic correlation between tasks is established. Combined with the dynamic correlation of these optimized task weights, the inter-task correlation matrix is ​​constructed in a weighted manner. The inter-task correlation matrix is ​​used to represent the mutual influence and feature sharing ratio between tasks. According to the inter-task correlation matrix, the information flow between tasks can be dynamically adjusted, so that different tasks can flexibly adjust the intensity and direction of feature sharing according to real-time training feedback, thereby maximizing the synergy between different tasks and improving the overall training performance.

[0059] Based on the information flow between tasks after dynamic adjustment, graph convolutional networks are used for modeling to improve the synergy between different tasks. It should be noted that by optimizing the synergy between tasks, each task can collaborate more efficiently when sharing features, avoiding situations where tasks are overly independent or overly dependent. Specifically, first, the convolutional layer and self-attention mechanism dynamically adjust the degree of dependence of each task on shared features by capturing the correlation and dependency between tasks, so that each task can flexibly adjust its feature sharing strategy based on real-time feedback. Subsequently, the graph convolutional network (GCN) further accurately models the information flow and interaction between tasks by constructing a dependency graph between tasks, thereby enhancing the efficiency of information transfer between tasks. This collaborative optimization enables different tasks to work more closely together during the training process, thereby improving the overall learning efficiency, maximizing the collaborative effect during the training process, and promoting the improvement of the overall training performance.

[0060] According to the training process of each task, the weights of different tasks in the joint feature extraction layer are adjusted through real-time feedback. The real-time feedback in the training process is combined with the feedback signal generated by virtual data to further optimize the joint feature extraction layer of the task.

[0061] S5: Monitor the patient's training status between different tasks in real time, perform posture corrections based on changes in movement patterns, and assess the degree of exercise fatigue to generate training feedback data.

[0062] Furthermore, the patient's exercise posture, exercise load and physiological state data in the training state between different tasks are monitored in real time, and Kalman filtering is used for denoising and time series alignment to output standardized training state data; Use temporal convolutional neural networks and long short-term memory networks to analyze standardized training state data and identify movement patterns and training task types in real time; It should be noted that movement patterns are defined as a combination of different movement patterns and postures exhibited by patients during rehabilitation training. For example, patients may perform movements such as knee flexion, standing, and gait. Different movement patterns can represent different stages or types of movements in the patient's rehabilitation process. Training task types refer to different training goals or task types set during the rehabilitation process. These task types are usually closely related to the patient's rehabilitation goals, and each task is trained for specific rehabilitation needs with the goal of helping patients restore motor function.

[0063] Based on the movement mode and training task type, the posture deviation in the movement mode is detected and corrected in real time; Specifically, according to the posture characteristics in the movement pattern and the training task objectives, the expected posture standard is first established and compared with the actual collected movement posture data to calculate the posture deviation. Then, the posture deviation is transmitted through the back propagation algorithm, and the real-time correction mechanism is used to adjust the parameters such as the movement load and movement angle in the training state to dynamically correct the posture deviation. In this process, the adjustment measures are matched with the patient's training task type to avoid adverse effects on the patient due to excessive load or excessive posture correction. Finally, the corrected posture data is combined with the updated training task type to adjust the training content to ensure that the patient maintains the best posture at each stage of rehabilitation training and optimize the training effect.

[0064] In traditional rehabilitation training monitoring methods, the detection and correction of posture deviations often rely on manual judgment or simple sensor feedback. This method has problems such as low recognition accuracy and poor real-time performance, making it difficult to achieve accurate personalized rehabilitation training. Especially in the process of critical rehabilitation, the patient's physiological state is complex and has significant individual differences, and traditional methods are difficult to cope with such high-complexity requirements.

[0065] Furthermore, by combining the movement mode and training task type, intelligent and real-time detection and correction of posture deviations are achieved, which specifically solves the real-time and personalization problems, as follows: Real-time problem: Through deep learning methods such as temporal convolutional neural network (TCN) and long short-term memory network (LSTM), dynamic features in motion data can be captured in real time, so that posture deviations can be instantly discovered and corrected during training. Personalization problem: Correction is carried out in combination with the patient's movement pattern and task type, avoiding the limitations of the "one-size-fits-all" approach in traditional solutions, so that each patient's rehabilitation training can be accurately adjusted according to their actual condition.

[0066] Frequency domain analysis is used to extract heart rate, electromyography, respiratory rate and blood oxygen saturation from physiological state data as physiological signals; Input physiological signals into a deep neural network to evaluate fatigue status, and adjust the exercise load in the training state according to the fatigue evaluation; It should be noted that the deep neural network extracts features from the patient's physiological signals and uses a multi-layer neural network model (such as convolutional neural network, long short-term memory network, etc.) to analyze the time domain and frequency domain features in the physiological signals to identify the patient's fatigue level. Specifically, the deep neural network automatically evaluates the patient's current fatigue state by comparing historical data and real-time data and analyzing the changing trend of physiological signals. Based on the fatigue state assessment, the exercise load in the training state is adjusted through an algorithm. The specific methods include automatically reducing or increasing the exercise intensity according to the degree of fatigue, adjusting the training duration or frequency, and optimizing the rest cycle, so as to ensure that the exercise load in the training state is within the patient's tolerance range. The entire process will provide real-time feedback on the fatigue state, and continuously adjust the training plan through continuous monitoring and evaluation to optimize the rehabilitation effect.

[0067] The adjusted training data is recorded in real time and integrated into training feedback data.

[0068] Among them, the integration step can be achieved by normalizing the adjusted training data recorded in real time, and then using data fusion algorithms (such as weighted averaging, feature splicing or principal component analysis, etc.) to merge multiple data types into unified training feedback data for subsequent analysis and cloud storage.

[0069] S6: Upload the training feedback data to the cloud platform for storage and analysis, generate a warning signal based on the cloud data analysis results, and automatically issue a warning when abnormalities occur in the patient's training.

[0070] Furthermore, the training feedback data is uploaded to the cloud platform through an encrypted channel for classified storage; Clean and denoise the classified and stored training feedback data, and extract the characteristics of exercise posture, exercise load and heart rate through deep learning analysis; The Kalman filter algorithm is used for cleaning and denoising to remove noise from the training feedback data and perform time series alignment to ensure the smoothness and accuracy of the data.

[0071] The cloud platform identifies and marks abnormal patterns in training feedback data based on exercise posture, exercise load and heart rate characteristics. Based on the abnormal patterns, the cloud generates warning signals and divides the warning levels according to the severity; Preferably, cloud analysis combined with deep learning methods can identify potential dangers such as excessive exercise load, posture deviation or abnormal physiological signals in real time. In particular, the comprehensive consideration of exercise load and physiological state can timely detect potential risks of patients during training, thereby issuing early warnings to prevent excessive exercise or improper posture from causing damage to patients' health.

[0072] According to the warning level, notifications are issued through multiple channels for real-time warning.

[0073] It should be noted that the warning level division is to solve the problem of different urgency of different types of abnormal events. On this basis, multi-channel notifications (such as APP push, text messages, etc.) are used to achieve real-time warnings to ensure that emergency situations at different levels are promptly fed back. Warnings in the prior art usually rely on only a single channel, while the present invention can more flexibly respond to different warning levels through the combination of multiple notification methods, so that relevant personnel (such as patients, trainers, and medical staff) can take effective measures in the shortest time. This multi-level and timely notification mechanism greatly improves practicality and response speed, and plays a key role in protecting the safety of patients.

[0074] The present embodiment also provides a critical care rehabilitation training monitoring device, including: a data collection module, a pattern generation module, a feature extraction module, a feature integration module, a state monitoring module and a cloud warning module; the data collection module is used to collect the patient's motion data in real time, preliminarily analyze the patient's current motion state, identify the patient's motion pattern, and output a preliminary motion pattern label; the pattern generation module is used to input the motion pattern label into the generative adversarial network, train the motion patterns and motion postures of different stages through the generator, and generate highly similar virtual data; the feature extraction module is used to initialize the motion data set through the virtual data, and input the motion data set into the multi-level convolutional neural network architecture for feature extraction , generate the patient's motor rehabilitation status; the feature integration module is used to learn the characteristics of different movement patterns through deep neural networks based on the patient's motor rehabilitation status, and integrate the characteristics of different movement patterns into a unified training network using a multi-task learning framework. The feature layer in the shared model is used to establish the correlation between different tasks; the state monitoring module is used to monitor the patient's training status between different tasks in real time, correct the posture of the change in the movement pattern, and evaluate the degree of fatigue of the movement to form training feedback data; the cloud warning module is used to upload the training feedback data to the cloud platform for storage and analysis, generate warning signals according to the cloud data analysis results, and automatically issue warnings when abnormalities occur in patient training.

[0075] This embodiment also provides a computer device, which is suitable for the case of a critical care rehabilitation training monitoring method, including: a memory and a processor; the memory is used to store computer executable instructions, and the processor is used to execute computer executable instructions to implement the critical care rehabilitation training monitoring method proposed in the above embodiment.

[0076] The computer device may be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device may be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device may be a touch layer covered on the display screen, or a key, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse, etc.

[0077] This embodiment also provides a storage medium on which a computer program is stored. When the program is executed by a processor, the method for implementing critical illness rehabilitation training monitoring as proposed in the above embodiment is implemented; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (Static Random Access Memory, referred to as SRAM), electrically erasable programmable read-only memory (Electrically Erasable Programmable Read-Only Memory, referred to as EEPROM), erasable programmable read-only memory (Erasable Programmable Read Only Memory, referred to as EPROM), programmable read-only memory (Programmable Red-Only Memory, referred to as PROM), read-only memory (Read-Only Memory, referred to as ROM), magnetic storage, flash memory, magnetic disk or optical disk.

[0078] In summary, the present invention can effectively improve the accuracy of motion pattern recognition and the evaluation effect of rehabilitation status through a multi-level deep learning architecture and a virtual data generation method. First, the dynamic features and temporal features in the motion data are extracted using a temporal convolutional neural network and a long short-term memory network, which can accurately capture the subtle changes in motion posture, motion load and rehabilitation status, thereby providing a more accurate motion pattern label. Secondly, by generating virtual data through a generative adversarial network and training in combination with a multi-level convolutional neural network, the training data set can be greatly enriched, the generalization ability can be improved, and the data scarcity problem that may occur during the actual data collection process can be overcome. In addition, based on a multi-task learning framework, the weights between different motion pattern features can be dynamically adjusted to enhance the synergy between tasks, thereby further improving the correction accuracy of the motion pattern and the real-time performance of fatigue assessment. Finally, through the intelligent early warning function of the cloud platform, an early warning signal can be issued in time when the patient's motion state is abnormal, which helps to realize a personalized rehabilitation training program.

[0079] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A critical care rehabilitation training monitoring method, characterized in that: include: Collect the patient's motion data in real time, preliminarily analyze the patient's current motion status, identify the patient's motion pattern, and output preliminary motion pattern labels; The motion pattern labels are input into the generative adversarial network, and the generator is used to train the motion patterns and motion postures at different stages to generate highly similar virtual data. Initialize the motion data set through virtual data, input the motion data set into a multi-level convolutional neural network architecture for feature extraction, and generate the patient's motor rehabilitation status; Based on the patient's motor rehabilitation status, the characteristics of different movement patterns are learned through deep neural networks. The multi-task learning framework is used to integrate the characteristics of different movement patterns into a unified training network. The feature layer in the shared model is used to establish the correlation between different tasks. Monitor the patient's training status between different tasks in real time, perform posture corrections based on changes in movement patterns, and assess the degree of fatigue during exercise to generate training feedback data; The training feedback data is uploaded to the cloud platform for storage and analysis, and a warning signal is generated based on the cloud data analysis results, and a warning is automatically issued when abnormalities occur in patient training.

2. The critical care rehabilitation training monitoring method according to claim 1, characterized in that: The motion data includes motion posture data, motion load data and motion state evaluation data; The preliminary analysis of the patient's current motion state, identification of the patient's motion pattern, and output of a preliminary motion pattern label are specifically performed as follows: The motion posture data, motion load data and motion state assessment data are denoised through timestamp alignment and Kalman filtering; Use a temporal convolutional neural network to extract dynamic features from the processed motion posture data and load features from the processed motion load data; Using deep neural networks to extract rehabilitation features from processed movement state assessment data; The extracted dynamic features, load features, and rehabilitation features are input into the long short-term memory network to capture the temporal features and identify the patient's movement pattern; The patient's movement pattern is corrected in combination with real-time movement status assessment data, and a preliminary movement pattern label is output.

3. The critical care rehabilitation training monitoring method according to claim 2, characterized in that: The motion pattern labels are input into the generative adversarial network, and the generator is used to train the motion patterns and motion postures at different stages to generate highly similar virtual data. The specific steps are as follows: The motion mode labels are used as input to the generator, where each label corresponds to a motion state in a time period; The generator uses a multi-layer structure that combines a bidirectional long short-term memory network with a self-attention mechanism to capture long-term and short-term dependencies in motion data. The different modal data in the motion data are mapped into a shared feature space through a cross-modal fusion layer, and the multi-layer structure of the generator is used to generate virtual data of motion modes at different stages. A deep convolutional neural network is used to build a discriminator to detect the authenticity of virtual data, and a temporal consistency loss is applied to ensure the temporal continuity of virtual data. Use virtual data generation models to supplement training datasets and pre-train the generator in combination with transfer learning methods; The motion pattern label is input into the pre-trained generator to generate virtual data, and after adversarial training of the discriminator, highly similar virtual data is output.

4. The critical care rehabilitation training monitoring method according to claim 3, characterized in that: The motion data set is initialized by virtual data, and the motion data set is input into a multi-level convolutional neural network architecture for feature extraction to generate the patient's motion rehabilitation status. The specific steps are as follows: The generated virtual data is standardized and amplified to construct a motion data set, and is labeled based on the automatic label generation method of deep learning to form a motion training set; The motion training set is input into a multi-layer convolutional neural network, and the primary and high-level representations of dynamic features, load features, and rehabilitation features are gradually extracted through the convolutional layers; The extracted primary and high-level representations are input into the fully connected layer for fusion, and the patient's motor rehabilitation status is generated through the classification and regression network.

5. The critical care rehabilitation training monitoring method according to claim 4, characterized in that: Based on the patient's motor rehabilitation status, the characteristics of different movement patterns are learned through deep neural networks, and the multi-task learning framework is used to integrate the characteristics of different movement patterns into a unified training network. The feature layer in the shared model is used to establish the correlation between different tasks. The specific steps are as follows: According to the patient's motor rehabilitation status, the first task is to identify the change pattern of the movement posture, the second task is to evaluate the change trend of the exercise load, and the third task is to evaluate the abnormal state recognition of the patient's rehabilitation status; After clarifying the task objectives, a joint feature extraction layer is constructed by jointly building a convolutional neural network and a deep neural network. In the joint feature extraction layer, the features of different motion modes are extracted by the convolutional layer in the initial stage, and then further fused by the fully connected layer to form a joint feature space. In the joint feature space, the joint feature extraction layer adjusts the feature sharing ratio between each task in real time through the weighted attention mechanism, thereby establishing dynamic correlation between different tasks; Based on the data features of each task, the degree of association between each task and the joint feature extraction layer is adaptively adjusted. In the multi-task learning framework, a weighted loss function is used to adjust the loss weight of each task, and the contribution between tasks is optimized based on feedback during training. Combine the adjusted loss weights between tasks to establish dynamic correlation, construct the correlation matrix between tasks, and dynamically adjust the information flow between tasks accordingly; Based on the information flow between dynamically adjusted tasks, graph convolutional networks are used for modeling; According to the training process of each task, the weights of different tasks in the joint feature extraction layer are adjusted through real-time feedback. The real-time feedback in the training process is combined with the feedback signal generated by virtual data to further optimize the joint feature extraction layer of the task.

6. The critical care rehabilitation training monitoring method according to claim 5, characterized in that: The real-time monitoring of the patient's training status between different tasks, the posture correction of the change of the movement pattern, and the evaluation of the degree of fatigue of the movement to form training feedback data are performed. The specific steps are as follows: Real-time monitoring of the patient's exercise posture, exercise load and physiological status data during training between different tasks, and denoising and time series alignment through Kalman filtering to output standardized training status data; Use temporal convolutional neural networks and long short-term memory networks to analyze standardized training state data and identify movement patterns and training task types in real time; Based on the movement mode and training task type, the posture deviation in the movement mode is detected and corrected in real time; Frequency domain analysis is used to extract heart rate, electromyography, respiratory rate and blood oxygen saturation from physiological state data as physiological signals; Input physiological signals into a deep neural network to evaluate fatigue status, and adjust the exercise load in the training state according to the fatigue evaluation; The adjusted training data is recorded in real time and integrated into training feedback data.

7. The critical care rehabilitation training monitoring method according to claim 6, characterized in that: The training feedback data is uploaded to the cloud platform for storage and analysis, an early warning signal is generated according to the cloud data analysis results, and an early warning is automatically issued when abnormalities occur in the patient's training. The specific steps are as follows: Upload the training feedback data to the cloud platform through an encrypted channel for classified storage; Clean and denoise the classified and stored training feedback data, and extract the characteristics of exercise posture, exercise load and heart rate through deep learning analysis; The cloud platform identifies and marks abnormal patterns in training feedback data based on exercise posture, exercise load and heart rate characteristics. Based on the abnormal patterns, the cloud generates warning signals and divides the warning levels according to the severity; According to the warning level, notifications are issued through multiple channels for real-time warning.

8. A critical care rehabilitation training monitoring device, based on the critical care rehabilitation training monitoring method according to any one of claims 1 to 7, characterized in that: It includes data collection module, pattern generation module, feature extraction module, feature integration module, status monitoring module and cloud warning module; The data collection module is used to collect the patient's motion data in real time, preliminarily analyze the patient's current motion state, identify the patient's motion pattern, and output a preliminary motion pattern label; The pattern generation module is used to input the motion pattern label into the generative adversarial network, and generate highly similar virtual data by training the motion patterns and motion postures at different stages through the generator; The feature extraction module is used to initialize the motion data set through virtual data, input the motion data set into a multi-level convolutional neural network architecture for feature extraction, and generate the patient's motor rehabilitation status; The feature integration module is used to learn the features of different movement patterns through a deep neural network based on the patient's motor rehabilitation status, integrate the features of different movement patterns into a unified training network using a multi-task learning framework, and establish the correlation between different tasks using the feature layer in the shared model; The state monitoring module is used to monitor the patient's training state between different tasks in real time, perform posture correction for changes in movement patterns, and evaluate the degree of fatigue during exercise to form training feedback data; The cloud-based early warning module is used to upload training feedback data to the cloud platform for storage and analysis, generate early warning signals based on cloud data analysis results, and automatically issue early warnings when abnormalities occur in patient training.

Images