Effective collection and analysis method of user characteristic data for lower limb hip and knee rehabilitation

Abstract

The present invention relates to the field of lower limb hip and knee data analysis, and discloses an effective collection and analysis method for user characteristic data for lower limb hip and knee rehabilitation, comprising the following steps: collecting relevant data of the user's lower limb hip and knee used for exercise and performing data preprocessing, while using a convolutional neural network to model and analyze the preprocessed relevant data, and designing and constructing a multi-channel sequence gait detection algorithm model to preset the user's lower limb hip and knee next action. Finally, the multi-channel sequence gait detection algorithm model is subjected to model loss training optimization, and a scoring mechanism is introduced, so that the multi-channel sequence gait detection algorithm model can perform gait scoring on the target user. The present invention can realize the simultaneous input of multiple channel data, realize the detection of the current state and gait of the human body, and can also achieve the effect of maintaining recognition in the absence of channel data without switching modes.

Description

Technical Field

[0001] The present invention relates to the field of lower limb hip and knee data analysis, and in particular to a method for effectively collecting and analyzing user characteristic data for lower limb hip and knee rehabilitation. Background Art

[0002] With the advent of an aging population, the number of elderly people in my country is increasing. By 2024, the number of people over 60 years old will be close to 300 million. People's demand for exercise and health is becoming increasingly strong. However, neurological diseases such as stroke are seriously affecting people's health. With the improvement of people's living standards, the incidence of stroke has further increased and is showing a trend of younger people. Therefore, the prevention and treatment of stroke and the rehabilitation of motor function after stroke are particularly important, especially the motor rehabilitation of the lower limbs, which is related to whether people can move freely and is more prominent. At the same time, trauma, disability and other problems can also lead to motor dysfunction, further aggravating people's demand for rehabilitation training. Traditional methods use manual techniques to achieve limb rehabilitation of patients, but traditional techniques are inefficient and have poor results. Therefore, rehabilitation robots have gradually emerged, with effects that are not inferior to manual ones, and are more efficient and comfortable.

[0003] A method for effectively collecting and analyzing user characteristic data for lower limb hip and knee rehabilitation is proposed. It is used to realize the simultaneous input of multiple channel data, realize the detection of the current state and gait of the human body, and can also adapt to the situation where channel data is missing and still maintain recognition without switching modes. Summary of the Invention

[0004] The present invention overcomes the deficiencies of the prior art and provides an effective method for collecting and analyzing user characteristic data for lower limb hip and knee rehabilitation.

[0005] In order to achieve the above object, the technical solution adopted by the present invention is:

[0006] A first aspect of the present invention provides a method for effectively collecting and analyzing user characteristic data for lower limb hip and knee rehabilitation, comprising the following steps:

[0007] Deploy sensors at the user's lower limb hips and knees, control the sensors to collect relevant data of the user's lower limb hips and knees, and preprocess the relevant data of the user's lower limb hips and knees to obtain preprocessed biocurrent values, preprocessed accelerations, preprocessed angular velocities, and preprocessed pressure data;

[0008] Based on convolutional neural networks, a multi-channel sequential gait detection algorithm model is designed by combining preprocessed biocurrent values, preprocessed acceleration, preprocessed angular velocity, and preprocessed pressure data.

[0009] The model loss training of the multi-channel sequence gait detection algorithm model is optimized, and a scoring mechanism is introduced to enable the multi-channel sequence gait detection algorithm model to score the gait of the target user.

[0010] Furthermore, in a preferred embodiment of the present invention, sensors are deployed at the hip and knee of the user's lower limbs, and the sensors are controlled to collect relevant data of the user's lower limbs, hip and knee. At the same time, the relevant data of the user's lower limbs, hip and knee are preprocessed to obtain preprocessed biocurrent value, preprocessed acceleration, preprocessed angular velocity and preprocessed pressure data, specifically:

[0011] The user who needs lower limb hip and knee rehabilitation is calibrated as the target user, and the location where the sensor needs to be deployed on the target user's lower limb hip and knee is determined and calibrated as the target deployment location, wherein the sensors include pressure sensors and surface electromyography sensors, and the sensor deployment locations include the target user's hip joint, knee joint and foot;

[0012] A pressure sensor and a surface electromyography sensor are respectively installed at the target deployment position, and a standard time for performing a lower limb hip and knee rehabilitation test is preset and calibrated as a target time;

[0013] During the target time, using a pressure sensor and a surface electromyography sensor, real-time pressure data and surface electromyography data are collected at the target deployment position of the target user. The pressure data includes pressure magnitude and pressure distribution, and the surface electromyography data includes biocurrent value, angular velocity, and acceleration of the muscle at the target deployment position.

[0014] When collecting pressure data and surface electromyography data at the target deployment position of the target user, different collection times of the pressure data and the surface electromyography data are simultaneously generated, and a collection timestamp is generated based on the different collection times of the pressure data and the surface electromyography data;

[0015] Based on the acquisition timestamp, the different pressure data and surface electromyography data collected are aligned for time consistency, and the acquisition frequencies of the pressure sensor and the surface electromyography sensor are controlled to be equal. At the same time, the collected pressure data and surface electromyography data are preprocessed and enhanced.

[0016] Furthermore, in a preferred embodiment of the present invention, the collected pressure data and surface electromyography data are subjected to data preprocessing and data enhancement processing, specifically:

[0017] The collected pressure data and surface electromyography data are collectively referred to as lower limb hip and knee data, and a Kalman filter algorithm is introduced to perform Kalman filtering on the lower limb hip and knee data to remove high-frequency noise from the lower limb hip and knee data to obtain lower limb hip and knee filtered data;

[0018] The lower limb hip and knee filter data are subjected to gravity compensation separation to obtain the biocurrent value, acceleration and angular velocity of the surface electromyography data in the lower limb hip and knee filter data, which are calibrated as preprocessed biocurrent value, preprocessed acceleration and preprocessed angular velocity. At the same time, the lower limb hip and knee filter data are pressure normalized to obtain preprocessed pressure data.

[0019] Furthermore, in a preferred embodiment of the present invention, the multi-channel sequential gait detection algorithm model is designed based on the convolutional neural network and combined with the pre-processed bioelectrical current value, pre-processed acceleration, pre-processed angular velocity and pre-processed pressure data, specifically:

[0020] Introducing a convolutional neural network and constructing a convolutional neural network blank model, wherein the convolutional neural network blank model includes a convolution layer, a pooling layer, and an output layer;

[0021] The preprocessed bioelectric current value, preprocessed acceleration, preprocessed angular velocity, and preprocessed pressure data are collectively referred to as target user feature data. The target user feature data are imported into the convolutional layer of a blank convolutional neural network model, and a multi-channel time series matrix is constructed for the target user feature data in the convolutional layer of the blank convolutional neural network model.

[0022] The multi-channel time series matrix of the target user feature data is a matrix for performing spatiotemporal alignment on the target user feature data;

[0023] In the convolution layer of the blank model of the convolutional neural network, the multi-channel time series matrix is traversed and convolved, and the time series lengths of the multi-channel time series matrix are kept equal during the traversal and convolution process;

[0024] After the convolution is completed, the multi-channel time series matrix after the convolution is imported into the pooling layer for global time series pooling to match the target user to adapt to different gaits. At the same time, a multi-channel attention mechanism is introduced into the blank model of the convolutional neural network during the global time series pooling process.

[0025] The multi-channel attention mechanism automatically focuses on the multi-channel time series matrix after convolution during the global time series pooling process to prevent data loss.

[0026] When the global time series pooling is completed, the trained multi-channel time series matrix is output at the output layer to obtain the multi-channel sequence gait detection algorithm model;

[0027] The multi-channel sequential gait detection algorithm model can predict the gait state of the target user within a specified time.

[0028] Furthermore, in a preferred embodiment of the present invention, the multi-channel sequence gait detection algorithm model is subjected to model loss training optimization, and a scoring mechanism is introduced to enable the multi-channel sequence gait detection algorithm model to score the gait of the target user, specifically:

[0029] Running a multi-channel sequential gait detection algorithm model, introducing smoothed cross entropy, and using the smoothed cross entropy as a loss function of the multi-channel sequential gait detection algorithm model, performing loss training on the multi-channel sequential gait detection algorithm model using the loss function, wherein the loss training is to use the smoothed cross entropy to perform category balancing training on the prediction results;

[0030] A cosine annealing algorithm is introduced. During loss training, the cosine annealing learning rate of the prediction result is calculated in real time by the cosine annealing algorithm, and a standard cosine annealing learning rate threshold is preset. If the cosine annealing learning rate of the prediction result reaches the standard cosine annealing learning rate threshold, the loss training is stopped. The prediction result after loss training is obtained and calibrated as a class of predicted user feature data. At the same time, the multi-channel sequence gait detection algorithm model after loss training is calibrated as the target multi-channel sequence gait detection algorithm model.

[0031] Determine the timestamp corresponding to a class of predicted user feature data, mark it as a class of timestamps, and collect the real feature data corresponding to the target user at the class of timestamps in real time, mark it as a class of real user feature data;

[0032] Among them, one type of predicted user feature data is the predicted bioelectrical current value, acceleration, angular velocity and pressure data of the target user's lower limb hip and knee within a specific time, and the other type of real user feature data is the real data of the bioelectrical current value, acceleration, angular velocity and pressure data of the target user's lower limb hip and knee within a specific time;

[0033] Calculating the Euclidean distance between a class of predicted user feature data and a class of real user feature data, calibrating it as a class of Euclidean distance, and classifying the target multi-channel sequential gait detection algorithm model as a qualified multi-channel sequential gait detection algorithm model if the class of Euclidean distance remains within a preset value;

[0034] If the Euclidean distance of a class does not maintain at the preset value, the target multi-channel sequence gait detection algorithm model is divided into the model to be analyzed, the model to be analyzed is optimized by edge computing, and a scoring mechanism is introduced so that the qualified multi-channel sequence gait detection algorithm model can score the gait of the target user.

[0035] Furthermore, in a preferred embodiment of the present invention, the edge computing optimization is performed on the model to be analyzed, and a scoring mechanism is introduced to enable the qualified multi-channel sequence gait detection algorithm model to perform gait scoring on the target user, specifically:

[0036] Performing lightweight edge computing processing on the model to be analyzed, wherein the lightweight edge computing processing of the model to be analyzed is to first perform channel pruning processing on the model to be analyzed, and then introduce a singular value decomposition algorithm to perform singular value decomposition on the model to be analyzed after the channel pruning processing, thereby reducing the computational complexity and memory usage of the model to be analyzed, and obtaining the target model to be analyzed;

[0037] Performing robustness verification on the target model to be analyzed, wherein the robustness verification is to partially mask the target user's feature data generated by the target model to be analyzed, and performing overlap analysis on the target user's feature data obtained after the partial masking and the complete data;

[0038] If the overlap is greater than the preset value, the target model to be analyzed is output and classified as a qualified multi-channel sequence gait detection algorithm model. If the overlap is not greater than the preset value, the target model to be analyzed continues to be subjected to model edge computing lightweight processing until a qualified multi-channel sequence gait detection algorithm model is generated.

[0039] A real-time scoring mechanism is introduced into the qualified multi-channel sequence gait detection algorithm model, and a user gait scoring table is constructed, in which scores corresponding to different user feature data are indicated;

[0040] The user feature data output by the qualified multi-channel sequence gait detection algorithm model is calibrated as qualified user feature data, and a gait state score is performed on the qualified user feature data based on the user gait scoring table.

[0041] A second aspect of the present invention further provides a system for effectively collecting and analyzing user characteristic data for lower limb hip and knee rehabilitation. The system includes a memory and a processor. The memory stores a method for effectively collecting and analyzing user characteristic data. When the method is executed by the processor, the following steps are implemented:

[0042] Deploy sensors at the user's lower limb hips and knees, control the sensors to collect relevant data of the user's lower limb hips and knees, and preprocess the relevant data of the user's lower limb hips and knees to obtain preprocessed biocurrent values, preprocessed accelerations, preprocessed angular velocities, and preprocessed pressure data;

[0043] Based on convolutional neural networks, a multi-channel sequential gait detection algorithm model is designed by combining preprocessed biocurrent values, preprocessed acceleration, preprocessed angular velocity, and preprocessed pressure data.

[0044] The model loss training of the multi-channel sequence gait detection algorithm model is optimized, and a scoring mechanism is introduced to enable the multi-channel sequence gait detection algorithm model to score the gait of the target user.

[0045] The present invention solves the technical defects existing in the background technology, and the present invention has the following beneficial effects: collecting relevant data of the user's hip and knee used for movement and performing data preprocessing, and using convolutional neural networks to model and analyze the preprocessed relevant data, and designing and constructing a multi-channel sequence gait detection algorithm model to preset the next action of the user's lower limb hip and knee. Finally, the multi-channel sequence gait detection algorithm model is trained and optimized for model loss, and a scoring mechanism is introduced to enable the multi-channel sequence gait detection algorithm model to score the gait of the target user. The present invention can realize the simultaneous input of multiple channel data, realize the detection of the current state and gait of the human body, and can also achieve the effect of maintaining recognition in the absence of channel data without switching modes. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, without paying any creative work, they can also obtain drawings of other embodiments based on these drawings.

[0047] Figure 1 A flow chart showing a method for effectively collecting and analyzing user characteristic data for lower limb hip and knee rehabilitation is provided;

[0048] Figure 2 A flow chart of a method for optimizing model loss training for a multi-channel sequential gait detection algorithm model is shown;

[0049] Figure 3 The program diagram of the system for effectively collecting and analyzing user characteristic data for lower limb hip and knee rehabilitation is shown. DETAILED DESCRIPTION

[0050] In order to more clearly understand the above-mentioned objects, features and advantages of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be noted that, in the absence of conflict, the embodiments of the present application and the features therein can be combined with each other.

[0051] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Therefore, the scope of protection of the present invention is not limited to the specific embodiments disclosed below.

[0052] Figure 1 A flow chart showing a method for effectively collecting and analyzing user characteristic data for lower limb hip and knee rehabilitation is provided, comprising the following steps:

[0053] S102: Deploy sensors at the hip and knee of the user's lower limbs, control the sensors to collect data related to the hip and knee of the user's lower limbs, and preprocess the data related to the hip and knee of the user's lower limbs to obtain preprocessed biocurrent value, preprocessed acceleration, preprocessed angular velocity, and preprocessed pressure data;

[0054] S104: Based on convolutional neural networks, combined with pre-processed bioelectrical current values, pre-processed acceleration, pre-processed angular velocity, and pre-processed pressure data, a multi-channel sequential gait detection algorithm model is designed;

[0055] S106: Perform model loss training optimization on the multi-channel sequence gait detection algorithm model, and introduce a scoring mechanism so that the multi-channel sequence gait detection algorithm model can score the gait of the target user.

[0056] Furthermore, in a preferred embodiment of the present invention, sensors are deployed at the hip and knee of the user's lower limbs, and the sensors are controlled to collect relevant data of the user's lower limbs, hip and knee. At the same time, the relevant data of the user's lower limbs, hip and knee are preprocessed to obtain preprocessed biocurrent value, preprocessed acceleration, preprocessed angular velocity and preprocessed pressure data, specifically:

[0057] The user who needs lower limb hip and knee rehabilitation is calibrated as the target user, and the location where the sensor needs to be deployed on the target user's lower limb hip and knee is determined and calibrated as the target deployment location, wherein the sensors include pressure sensors and surface electromyography sensors, and the sensor deployment locations include the target user's hip joint, knee joint and foot;

[0058] A pressure sensor and a surface electromyography sensor are respectively installed at the target deployment position, and a standard time for performing a lower limb hip and knee rehabilitation test is preset and calibrated as a target time;

[0059] During the target time, using a pressure sensor and a surface electromyography sensor, real-time pressure data and surface electromyography data are collected at the target deployment position of the target user. The pressure data includes pressure magnitude and pressure distribution, and the surface electromyography data includes biocurrent value, angular velocity, and acceleration of the muscle at the target deployment position.

[0060] When collecting pressure data and surface electromyography data at the target deployment position of the target user, different collection times of the pressure data and the surface electromyography data are simultaneously generated, and a collection timestamp is generated based on the different collection times of the pressure data and the surface electromyography data;

[0061] Based on the acquisition timestamp, the different pressure data and surface electromyography data collected are aligned for time consistency, and the acquisition frequencies of the pressure sensor and the surface electromyography sensor are controlled to be equal. At the same time, the collected pressure data and surface electromyography data are preprocessed and enhanced.

[0062] It should be noted that lower limb hip and knee rehabilitation requires the assistance of a robotic arm, which in turn requires the collection of human data to facilitate lower limb hip and knee movement rehabilitation. Human data is collected and predicted, predicting the body's next movement. The robotic arm then drives lower limb hip and knee movement, achieving assisted rehabilitation. First, data is collected from the hip and knee joints, as these are the areas that propel the body forward. Pressure sensors and surface electromyography sensors are installed for data collection. Surface electromyography sensors are attached to key muscle groups, such as the rectus femoris and biceps femoris, to capture muscle activation timing and collect bioelectrical current values, angular velocity, and acceleration at the target muscle locations, essential for prediction. Pressure data is recorded by embedded insoles or ground force plates, recording plantar pressure distribution. All collected data is time-synchronized to ensure consistency and authenticity during data transmission and analysis. The sampling frequency must be consistent to prevent constant time synchronization adjustments.

[0063] Furthermore, in a preferred embodiment of the present invention, the collected pressure data and surface electromyography data are subjected to data preprocessing and data enhancement processing, specifically:

[0064] The collected pressure data and surface electromyography data are collectively referred to as lower limb hip and knee data, and a Kalman filter algorithm is introduced to perform Kalman filtering on the lower limb hip and knee data to remove high-frequency noise from the lower limb hip and knee data to obtain lower limb hip and knee filtered data;

[0065] The lower limb hip and knee filter data are subjected to gravity compensation separation to obtain the biocurrent value, acceleration and angular velocity of the surface electromyography data in the lower limb hip and knee filter data, which are calibrated as preprocessed biocurrent value, preprocessed acceleration and preprocessed angular velocity. At the same time, the lower limb hip and knee filter data are pressure normalized to obtain preprocessed pressure data.

[0066] It should be noted that the purpose of filtering and enhancing the lower limb hip and knee data is to reduce the error rate and improve accuracy during the modeling process. The Kalman filter algorithm can remove high-frequency noise from data, but the filtered data may contain intersections, requiring data redistribution. Therefore, gravity compensation separation is used to separate the biocurrent value, acceleration, and angular velocity of the surface electromyography data.

[0067] Furthermore, in a preferred embodiment of the present invention, the multi-channel sequential gait detection algorithm model is designed based on the convolutional neural network and combined with the pre-processed bioelectrical current value, pre-processed acceleration, pre-processed angular velocity and pre-processed pressure data, specifically:

[0068] Introducing a convolutional neural network and constructing a convolutional neural network blank model, wherein the convolutional neural network blank model includes a convolution layer, a pooling layer, and an output layer;

[0069] The preprocessed bioelectric current value, preprocessed acceleration, preprocessed angular velocity, and preprocessed pressure data are collectively referred to as target user feature data. The target user feature data are imported into the convolutional layer of a blank convolutional neural network model, and a multi-channel time series matrix is constructed for the target user feature data in the convolutional layer of the blank convolutional neural network model.

[0070] The multi-channel time series matrix of the target user feature data is a matrix for performing spatiotemporal alignment on the target user feature data;

[0071] In the convolution layer of the blank model of the convolutional neural network, the multi-channel time series matrix is traversed and convolved, and the time series lengths of the multi-channel time series matrix are kept equal during the traversal and convolution process;

[0072] After the convolution is completed, the multi-channel time series matrix after the convolution is imported into the pooling layer for global time series pooling to match the target user to adapt to different gaits. At the same time, a multi-channel attention mechanism is introduced into the blank model of the convolutional neural network during the global time series pooling process.

[0073] The multi-channel attention mechanism automatically focuses on the multi-channel time series matrix after convolution during the global time series pooling process to prevent data loss.

[0074] When the global time series pooling is completed, the trained multi-channel time series matrix is output at the output layer to obtain the multi-channel sequence gait detection algorithm model;

[0075] The multi-channel sequential gait detection algorithm model can predict the gait state of the target user within a specified time.

[0076] It should be noted that the multi-channel sequential gait detection algorithm model can simultaneously process the user's lower limb hip and knee data from multiple channels, enabling detection of the body's current state and gait. It can also adapt to missing channel data and maintain recognition without switching modes. A convolutional neural network is a prediction algorithm used to construct the multi-channel sequential gait detection model. The multi-channel sequential gait detection model consists of convolutional and pooling layers. The convolutional layers are training layers used to train preprocessed biocurrent values, preprocessed acceleration, preprocessed angular velocity, and preprocessed pressure data to determine the user's gait over the next period of time. The multi-channel time series matrix is constructed by the convolutional layers and is a key component within the convolutional layers. It contains all preprocessed biocurrent values, preprocessed acceleration, preprocessed angular velocity, and preprocessed pressure data. Maintaining the same time series length in the multi-channel time series matrix ensures that the pooling process does not cause errors. Global time series pooling is used to adapt to the different gaits and speeds of the target users. A multi-channel attention mechanism is introduced to prevent data loss and automatically focus on valid information.

[0077] Figure 2 A flow chart of a method for optimizing model loss training for a multi-channel sequential gait detection algorithm model is shown, including the following steps:

[0078] S202: Optimizing model loss training for the multi-channel sequence gait detection algorithm model, and introducing a scoring mechanism so that the multi-channel sequence gait detection algorithm model can score the gait of the target user;

[0079] S204: Perform edge computing optimization on the model to be analyzed and introduce a scoring mechanism so that the qualified multi-channel sequence gait detection algorithm model can perform gait scoring on the target user.

[0080] Furthermore, in a preferred embodiment of the present invention, the multi-channel sequence gait detection algorithm model is subjected to model loss training optimization, and a scoring mechanism is introduced to enable the multi-channel sequence gait detection algorithm model to score the gait of the target user, specifically:

[0081] Running a multi-channel sequential gait detection algorithm model, introducing smoothed cross entropy, and using the smoothed cross entropy as a loss function of the multi-channel sequential gait detection algorithm model, performing loss training on the multi-channel sequential gait detection algorithm model using the loss function, wherein the loss training is to use the smoothed cross entropy to perform category balancing training on the prediction results;

[0082] A cosine annealing algorithm is introduced. During loss training, the cosine annealing learning rate of the prediction result is calculated in real time by the cosine annealing algorithm, and a standard cosine annealing learning rate threshold is preset. If the cosine annealing learning rate of the prediction result reaches the standard cosine annealing learning rate threshold, the loss training is stopped. The prediction result after loss training is obtained and calibrated as a class of predicted user feature data. At the same time, the multi-channel sequence gait detection algorithm model after loss training is calibrated as the target multi-channel sequence gait detection algorithm model.

[0083] Determine the timestamp corresponding to a class of predicted user feature data, mark it as a class of timestamps, and collect the real feature data corresponding to the target user at the class of timestamps in real time, mark it as a class of real user feature data;

[0084] Among them, one type of predicted user feature data is the predicted bioelectrical current value, acceleration, angular velocity and pressure data of the target user's lower limb hip and knee within a specific time, and the other type of real user feature data is the real data of the bioelectrical current value, acceleration, angular velocity and pressure data of the target user's lower limb hip and knee within a specific time;

[0085] Calculating the Euclidean distance between a class of predicted user feature data and a class of real user feature data, calibrating it as a class of Euclidean distance, and classifying the target multi-channel sequential gait detection algorithm model as a qualified multi-channel sequential gait detection algorithm model if the class of Euclidean distance remains within a preset value;

[0086] If the Euclidean distance of a class does not maintain at the preset value, the target multi-channel sequence gait detection algorithm model is divided into the model to be analyzed, the model to be analyzed is optimized by edge computing, and a scoring mechanism is introduced so that the qualified multi-channel sequence gait detection algorithm model can score the gait of the target user.

[0087] It should be noted that smoothed cross entropy is a loss function used to mitigate the class imbalance of all predicted data and achieve the goal of smoothing the data to produce a true output. The cosine annealing learning rate represents the accuracy of the predicted data. When the cosine annealing learning rate reaches the standard value, it proves that the accuracy of the predicted data has reached the standard value. At this time, the target multi-channel sequence gait detection algorithm model is output. The predicted data obtained by the target multi-channel sequence gait detection algorithm model, that is, the feature data predicted by the user at the lower limb hip and knee, needs to be verified for authenticity. Authenticity can be determined by the data similarity between the actual value and the predicted value. The Euclidean distance inference method is an algorithm for determining data similarity. When the Euclidean distance remains within the standard value, it proves that the data similarity is high, which also proves that the data predicted by the target multi-channel sequence gait detection algorithm model is authentic and effective and can be used for lower limb hip and knee rehabilitation. If the Euclidean distance does not remain within the standard value, the model needs to be optimized.

[0088] Furthermore, in a preferred embodiment of the present invention, the edge computing optimization is performed on the model to be analyzed, and a scoring mechanism is introduced to enable the qualified multi-channel sequence gait detection algorithm model to perform gait scoring on the target user, specifically:

[0089] Performing lightweight edge computing processing on the model to be analyzed, wherein the lightweight edge computing processing of the model to be analyzed is to first perform channel pruning processing on the model to be analyzed, and then introduce a singular value decomposition algorithm to perform singular value decomposition on the model to be analyzed after the channel pruning processing, thereby reducing the computational complexity and memory usage of the model to be analyzed, and obtaining the target model to be analyzed;

[0090] Performing robustness verification on the target model to be analyzed, wherein the robustness verification is to partially mask the target user's feature data generated by the target model to be analyzed, and performing overlap analysis on the target user's feature data obtained after the partial masking and the complete data;

[0091] If the overlap is greater than the preset value, the target model to be analyzed is output and classified as a qualified multi-channel sequence gait detection algorithm model. If the overlap is not greater than the preset value, the target model to be analyzed continues to be subjected to model edge computing lightweight processing until a qualified multi-channel sequence gait detection algorithm model is generated.

[0092] A real-time scoring mechanism is introduced into the qualified multi-channel sequence gait detection algorithm model, and a user gait scoring table is constructed, in which scores corresponding to different user feature data are indicated;

[0093] The user feature data output by the qualified multi-channel sequence gait detection algorithm model is calibrated as qualified user feature data, and a gait state score is performed on the qualified user feature data based on the user gait scoring table.

[0094] It should be noted that edge computing optimization of the model involves lightweighting the model, which can reduce its computational complexity and improve its accuracy. Pruning removes redundant data from the model, while the singular value decomposition algorithm reduces computational complexity. The target model obtained after edge computing optimization undergoes robustness verification. This is done by manually randomly masking approximately 30% of the channel data and comparing the overlap with the complete data. If robustness is verified within a preset range, the model is considered a qualified multi-channel sequential gait detection algorithm model. If it fails, lightweighting is continued until it passes. The real-time scoring mechanism introduced in the qualified multi-channel sequential gait detection algorithm model is designed to ensure that if the next movement predicted by the hip and knee system meets the gait score, the user can perform the rehabilitation without intervention, allowing the user to perform the movement themselves, resulting in more effective rehabilitation. If the predicted gait score for the next movement fails, this indicates that the user's hip and knee will have difficulty performing the next movement and may require robotic assistance. Therefore, the gait score is used to determine whether robotic intervention is warranted.

[0095] like Figure 3 As shown, the second aspect of the present invention further provides a user characteristic data effective collection and analysis system for lower limb hip and knee rehabilitation, the user characteristic data effective collection and analysis system includes a memory 31 and a processor 32, the memory 31 stores a user characteristic data effective collection and analysis method, and when the user characteristic data effective collection and analysis method is executed by the processor 32, the following steps are implemented:

[0096] Deploy sensors at the user's lower limb hips and knees, control the sensors to collect relevant data of the user's lower limb hips and knees, and preprocess the relevant data of the user's lower limb hips and knees to obtain preprocessed biocurrent values, preprocessed accelerations, preprocessed angular velocities, and preprocessed pressure data;

[0097] Based on convolutional neural networks, a multi-channel sequential gait detection algorithm model is designed by combining preprocessed biocurrent values, preprocessed acceleration, preprocessed angular velocity, and preprocessed pressure data.

[0098] The model loss training of the multi-channel sequence gait detection algorithm model is optimized, and a scoring mechanism is introduced to enable the multi-channel sequence gait detection algorithm model to score the gait of the target user.

[0099] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in the present invention should be included in the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be based on the scope of protection of the claims.

Claims

1. A method for effectively collecting and analyzing user characteristic data for lower limb hip and knee rehabilitation, characterized in that: The following steps are involved: Deploy sensors at the user's lower limb hips and knees, and control the sensors to collect relevant data of the user's lower limb hips and knees. At the same time, preprocess the relevant data of the user's lower limb hips and knees to obtain preprocessed biocurrent values, preprocessed accelerations, preprocessed angular velocities, and preprocessed pressure data; Based on convolutional neural networks, a multi-channel sequential gait detection algorithm model is designed by combining preprocessed biocurrent values, preprocessed acceleration, preprocessed angular velocity, and preprocessed pressure data. The multi-channel sequence gait detection algorithm model is trained and optimized for model loss. At the same time, a scoring mechanism is introduced to enable the multi-channel sequence gait detection algorithm model to score the gait of the target user. Among them, based on the convolutional neural network, combined with pre-processed bioelectrical current values, pre-processed acceleration, pre-processed angular velocity and pre-processed pressure data, a multi-channel sequential gait detection algorithm model is designed, specifically: Introducing a convolutional neural network and constructing a convolutional neural network blank model, wherein the convolutional neural network blank model includes a convolution layer, a pooling layer, and an output layer; The preprocessed bioelectric current value, preprocessed acceleration, preprocessed angular velocity, and preprocessed pressure data are collectively referred to as target user feature data. The target user feature data are imported into the convolutional layer of a blank convolutional neural network model, and a multi-channel time series matrix is constructed for the target user feature data in the convolutional layer of the blank convolutional neural network model. The multi-channel time series matrix of the target user feature data is a matrix for performing spatiotemporal alignment on the target user feature data; In the convolution layer of the blank model of the convolutional neural network, the multi-channel time series matrix is traversed and convolved, and the time series lengths of the multi-channel time series matrix are kept equal during the traversal and convolution process; After the convolution is completed, the multi-channel time series matrix after the convolution is imported into the pooling layer for global time series pooling to match the target user to adapt to different gaits. At the same time, a multi-channel attention mechanism is introduced into the blank model of the convolutional neural network during the global time series pooling process. The multi-channel attention mechanism automatically focuses on the multi-channel time series matrix after convolution during the global time series pooling process to prevent data loss. When the global time series pooling is completed, the trained multi-channel time series matrix is output at the output layer to obtain the multi-channel sequence gait detection algorithm model; The multi-channel sequential gait detection algorithm model can predict the gait state of the target user within a specified time.

2. The method for effectively collecting and analyzing user characteristic data for lower limb hip and knee rehabilitation according to claim 1, characterized in that: The sensors are deployed at the hip and knee of the user's lower limbs, and the sensors are controlled to collect relevant data of the user's lower limbs, hip and knee. At the same time, the relevant data of the user's lower limbs, hip and knee are preprocessed to obtain preprocessed biocurrent value, preprocessed acceleration, preprocessed angular velocity and preprocessed pressure data, specifically: The user who needs lower limb hip and knee rehabilitation is calibrated as the target user, and the location where the sensor needs to be deployed on the target user's lower limb hip and knee is determined and calibrated as the target deployment location, wherein the sensors include pressure sensors and surface electromyography sensors, and the sensor deployment locations include the target user's hip joint, knee joint and foot; A pressure sensor and a surface electromyography sensor are respectively installed at the target deployment position, and a standard time for performing a lower limb hip and knee rehabilitation test is preset and calibrated as a target time; During the target time, using a pressure sensor and a surface electromyography sensor, real-time pressure data and surface electromyography data are collected at the target deployment position of the target user. The pressure data includes pressure magnitude and pressure distribution, and the surface electromyography data includes biocurrent value, angular velocity, and acceleration of the muscle at the target deployment position. When collecting pressure data and surface electromyography data at the target deployment position of the target user, different collection times of the pressure data and the surface electromyography data are simultaneously generated, and a collection timestamp is generated based on the different collection times of the pressure data and the surface electromyography data; Based on the acquisition timestamp, the different pressure data and surface electromyography data collected are aligned for time consistency, and the acquisition frequencies of the pressure sensor and the surface electromyography sensor are controlled to be equal. At the same time, the collected pressure data and surface electromyography data are preprocessed and enhanced.

3. The method for effectively collecting and analyzing user characteristic data for lower limb hip and knee rehabilitation according to claim 2, characterized in that: The data preprocessing and data enhancement processing of the collected pressure data and surface electromyography data are specifically as follows: The collected pressure data and surface electromyography data are collectively referred to as lower limb hip and knee data, and a Kalman filter algorithm is introduced to perform Kalman filtering on the lower limb hip and knee data to remove high-frequency noise from the lower limb hip and knee data to obtain lower limb hip and knee filtered data; The lower limb hip and knee filter data are subjected to gravity compensation separation to obtain the biocurrent value, acceleration and angular velocity of the surface electromyography data in the lower limb hip and knee filter data, which are calibrated as preprocessed biocurrent value, preprocessed acceleration and preprocessed angular velocity. At the same time, the lower limb hip and knee filter data are pressure normalized to obtain preprocessed pressure data.

4. The method for effectively collecting and analyzing user characteristic data for lower limb hip and knee rehabilitation according to claim 1, characterized in that: The multi-channel sequence gait detection algorithm model is trained and optimized for model loss, and a scoring mechanism is introduced to enable the multi-channel sequence gait detection algorithm model to score the gait of the target user, specifically: Running a multi-channel sequential gait detection algorithm model, introducing smoothed cross entropy, and using the smoothed cross entropy as a loss function of the multi-channel sequential gait detection algorithm model, performing loss training on the multi-channel sequential gait detection algorithm model using the loss function, wherein the loss training is to use the smoothed cross entropy to perform category balancing training on the prediction results; A cosine annealing algorithm is introduced. During loss training, the cosine annealing learning rate of the prediction result is calculated in real time by the cosine annealing algorithm, and a standard cosine annealing learning rate threshold is preset. If the cosine annealing learning rate of the prediction result reaches the standard cosine annealing learning rate threshold, the loss training is stopped. The prediction result after loss training is obtained and calibrated as a class of predicted user feature data. At the same time, the multi-channel sequence gait detection algorithm model after loss training is calibrated as the target multi-channel sequence gait detection algorithm model. Determine the timestamp corresponding to a class of predicted user feature data, mark it as a class of timestamps, and collect the real feature data corresponding to the target user at the class of timestamps in real time, mark it as a class of real user feature data; Among them, one type of predicted user feature data is the predicted bioelectrical current value, acceleration, angular velocity and pressure data of the target user's lower limb hip and knee within a specific time, and the other type of real user feature data is the real data of the bioelectrical current value, acceleration, angular velocity and pressure data of the target user's lower limb hip and knee within a specific time; Calculating the Euclidean distance between a class of predicted user feature data and a class of real user feature data, calibrating it as a class of Euclidean distance, and classifying the target multi-channel sequential gait detection algorithm model as a qualified multi-channel sequential gait detection algorithm model if the class of Euclidean distance remains within a preset value; If the Euclidean distance of a class does not maintain at the preset value, the target multi-channel sequence gait detection algorithm model is divided into the model to be analyzed, the model to be analyzed is optimized by edge computing, and a scoring mechanism is introduced so that the qualified multi-channel sequence gait detection algorithm model can score the gait of the target user.

5. The method for effectively collecting and analyzing user characteristic data for lower limb hip and knee rehabilitation according to claim 4, characterized in that: The edge computing optimization of the model to be analyzed and the introduction of a scoring mechanism enable the qualified multi-channel sequence gait detection algorithm model to score the gait of the target user, specifically: Performing lightweight edge computing processing on the model to be analyzed, wherein the lightweight edge computing processing of the model to be analyzed is to first perform channel pruning processing on the model to be analyzed, and then introduce a singular value decomposition algorithm to perform singular value decomposition on the model to be analyzed after the channel pruning processing, thereby reducing the computational complexity and memory usage of the model to be analyzed, and obtaining the target model to be analyzed; Performing robustness verification on the target model to be analyzed, wherein the robustness verification is to partially mask the target user's feature data generated by the target model to be analyzed, and performing overlap analysis on the target user's feature data obtained after the partial masking and the complete data; If the overlap is greater than the preset value, the target model to be analyzed is output and classified as a qualified multi-channel sequence gait detection algorithm model. If the overlap is not greater than the preset value, the target model to be analyzed continues to be subjected to model edge computing lightweight processing until a qualified multi-channel sequence gait detection algorithm model is generated. A real-time scoring mechanism is introduced into the qualified multi-channel sequence gait detection algorithm model, and a user gait scoring table is constructed, in which scores corresponding to different user feature data are indicated; The user feature data output by the qualified multi-channel sequence gait detection algorithm model is calibrated as qualified user feature data, and a gait state score is performed on the qualified user feature data based on the user gait scoring table.

6. A user characteristic data effective collection and analysis system for lower limb hip and knee rehabilitation, characterized by: The user feature data effective collection and analysis system includes a memory and a processor, and the memory stores a user feature data effective collection and analysis method program. When the user feature data effective collection and analysis method program is executed by the processor, the user feature data effective collection and analysis method steps as described in any one of claims 1 to 5 are implemented.

Images