Lower limb torque determination system, method, apparatus, computer equipment and storage medium

Abstract

This invention discloses a lower limb torque determination system, method, apparatus, computer device, and storage medium, comprising: acquiring a plantar pressure dataset, wherein the data in the plantar pressure dataset is time-series data; employing a multi-scale convolutional neural network to extract hierarchical features from the plantar pressure dataset, capturing local, intermediate, and global features within the hierarchical features to reflect the dynamic changes of plantar pressure at different levels; using an attention-enhanced graph convolutional network to extract time-series features at different levels of the plantar pressure dataset and dynamically assigning weights to key nodes to optimize the expression of important features; and employing a multi-source fusion and residual connection method to integrate hierarchical features, time-series features, and key node weights into a unified feature vector to calculate the lower limb torque prediction result. This invention improves the accuracy and efficiency of lower limb torque calculation in dynamic scenarios.

Description

Technical Field

[0001] This invention relates to the fields of biomechanics and sensor technology, and in particular to a lower limb torque determination system, method, device, computer equipment, and storage medium. Background Technology

[0002] In the field of biomechanics, accurate calculation of lower limb joint torques is of great significance for motion monitoring, rehabilitation therapy, and exoskeleton robotics. Traditional methods for determining lower limb torques employ inverse dynamics models, utilizing motion capture systems and ground reaction force platforms to acquire lower limb motion trajectories and mechanical data, thereby calculating the lower limb torques. While this method offers high computational accuracy, it also has several limitations, such as high equipment costs, stringent experimental requirements, and difficulty in adapting to outdoor or dynamic scenarios.

[0003] In existing methods, Chinese patent application CN118003341A discloses a method for calculating lower limb joint torque based on reinforcement learning agents. This method requires at least seven high-speed infrared cameras and a ground reaction force platform, costing over 2 million RMB. The high-speed infrared cameras have a sampling frequency of at least 200Hz, costing at least an order of magnitude more than ordinary cameras (typically 50Hz to 60Hz). The camera area requires constant illumination, which is difficult to achieve in outdoor experimental settings. The ground reaction force platform consists of a matrix of multi-dimensional force sensors, which have strict requirements for ground flatness and hardness, typically limiting its use to laboratory settings. Furthermore, the system requires strict synchronization of multiple devices, further increasing its complexity.

[0004] Chinese patent application CN117764151A discloses a method for predicting lower limb joint torque based on deep forest and genetic algorithm. This method improves computational accuracy by optimizing sensor layout, but its fitness evaluation model relies too heavily on static scenes and does not fully consider the impact of sensor signal fluctuations on torque calculation under dynamic environments. In addition, the complex multi-granularity scanning and cascaded forest model result in low computational efficiency, making it difficult to meet the requirements of real-time monitoring.

[0005] In the process of realizing this invention, the inventors realized that the prior art has at least the following technical problems: the existing methods have high requirements for equipment cost and experimental environment, which limits their applicability in outdoor or dynamic scenarios; the algorithm has limited ability in multi-source data fusion and feature extraction, and it is difficult to balance real-time performance and calculation accuracy, resulting in low economic cost and low compatibility of lower limb torque calculation. Summary of the Invention

[0006] This invention provides a lower limb torque determination system, method, apparatus, computer device, and storage medium to improve the accuracy and efficiency of lower limb torque calculation.

[0007] To address the aforementioned technical problems, this application provides a lower limb torque determination system, including a sensor module, a data acquisition module, and a processing module, wherein the sensor module, the data acquisition module, and the processing module are electrically connected.

[0008] The sensor module acquires data through various sensors and transmits the acquired data to the data acquisition module. The sensor module includes a plantar pressure controller and an IMU sensor. The plantar pressure controller is placed on the bottom of the wearable device or shoe to collect the contact force between the sole of the foot and the ground. The IMU sensor is sparsely arranged and fixed at the joint of the lower limb to collect multidimensional acceleration and angular velocity data to reduce the complexity of sensor deployment and power consumption.

[0009] The data acquisition module receives the data transmitted by the sensor module, performs standardized processing on the data, and then synchronously transmits it to the processing module. The data acquisition module includes an STM32F105 microcontroller and a Bluetooth device, which are used for data protocol conversion, high-precision time synchronization, adaptive data transmission, as well as data verification and automatic retransmission tasks.

[0010] The processing module receives data from the data acquisition module and determines the lower limb torque based on the received data.

[0011] Optionally, the Bluetooth device employs cyclic redundancy check to identify data transmission errors and, combined with a data retransmission strategy, improves data transmission reliability.

[0012] Optionally, the STM32F105 microcontroller interacts and synchronously collects data with the plantar pressure controller and the IMU sensor, and supports high-precision timestamp recording to ensure the consistency and integrity of the timing data.

[0013] To address the aforementioned technical problems, this application also provides a method for determining lower limb torque, comprising:

[0014] Obtain a plantar pressure dataset, wherein the data in the plantar pressure dataset is time-series data;

[0015] A multi-scale convolutional neural network is used to extract hierarchical features from the plantar pressure dataset. The hierarchical features include local features, intermediate features, and global features, which reflect the dynamic changes of plantar pressure at different levels.

[0016] Based on an attention-enhanced graph convolutional network, features are extracted from the plantar pressure dataset to obtain temporal features at different levels. The temporal features are then combined with attention to determine the weights of key nodes.

[0017] A multi-source fusion method combining feature concatenation and weighted averaging is adopted to integrate hierarchical features, temporal features, and key node weights into a unified feature vector. The original feature information is preserved through residual connection, reducing information loss during feature processing, and the lower limb torque prediction result is calculated.

[0018] Optionally, obtaining the plantar pressure dataset includes:

[0019] Receive real-time plantar pressure data, which includes plantar pressure, multidimensional acceleration, and angular velocity;

[0020] Add a timestamp to each of the real-time plantar pressure data points to obtain preliminary time-series data;

[0021] An adaptive Kalman filter algorithm is used to correct and denoise the preliminary time-series data to obtain the plantar pressure dataset.

[0022] Optionally, determining the key node weights by combining the temporal features with attention includes:

[0023] Using an attention mechanism and the aforementioned temporal features, the contribution value of each node is calculated to obtain the feature importance weight value of each node.

[0024] By combining neighborhood information, the relationships between different sections can be captured;

[0025] The weights of key nodes are determined based on the feature importance weights of each node and the correlations between nodes.

[0026] Optionally, a multi-source fusion method combining feature concatenation and weighted averaging is adopted to integrate hierarchical features, temporal features, and key node weights into a unified feature vector; after the step of retaining the original feature information through residual connection to reduce information loss during feature processing and calculating the lower limb torque prediction result, the lower limb torque determination method further includes:

[0027] The predicted lower limb torque is evaluated to obtain the evaluation result;

[0028] Based on the evaluation results, a Bayesian optimization mechanism is used to dynamically adjust the feature weight allocation strategy to further improve the model's predictive performance.

[0029] To address the aforementioned technical problems, this application also provides a lower limb torque determination device, comprising:

[0030] The data acquisition module is used to acquire plantar pressure dataset, wherein the data in the plantar pressure dataset is time-series data;

[0031] The first extraction module is used to extract hierarchical features from the plantar pressure dataset using a multi-scale convolutional neural network. The hierarchical features include local features, intermediate features, and global features.

[0032] The second extraction module is used to extract features from the plantar pressure dataset based on an attention-enhanced graph convolutional network to obtain temporal features at different levels, and to use the temporal features in combination with attention to determine the weights of key nodes.

[0033] The fusion calculation module adopts a multi-source fusion method that combines feature splicing and weighted averaging to integrate hierarchical features, temporal features, and key node weights into a unified feature vector; it retains the original feature information through residual connection, reduces information loss during feature processing, and calculates the lower limb torque prediction result.

[0034] Optionally, the data acquisition module includes:

[0035] A data receiving unit is used to receive real-time plantar pressure data, which includes plantar pressure, multidimensional acceleration, and angular velocity.

[0036] The timestamp addition unit is used to add a timestamp to each of the real-time plantar pressure data to obtain preliminary time-series data.

[0037] The data preprocessing unit is used to perform data correction and noise reduction on the preliminary time series data using an adaptive Kalman filter algorithm to obtain the plantar pressure dataset.

[0038] Optionally, the second extraction module includes:

[0039] The contribution calculation unit is used to calculate the contribution value of each node using the attention mechanism and the temporal features, so as to obtain the feature importance weight value of each node.

[0040] The association computing unit is used to combine neighborhood information to capture the dynamic interaction characteristics between nodes;

[0041] The weight generation unit is used to determine the weight of key nodes based on the feature importance weight value of each node and the correlation between nodes.

[0042] Optionally, the lower limb torque determining device further includes:

[0043] The result evaluation module is used to evaluate the lower limb torque prediction results and obtain the evaluation results.

[0044] The weight update module is used to dynamically adjust the feature weight allocation strategy based on the evaluation results using a Bayesian optimization mechanism, so as to further improve the model prediction performance.

[0045] To address the aforementioned technical problems, this application also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the aforementioned lower limb torque determination method.

[0046] To address the aforementioned technical problems, this application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the aforementioned lower limb torque determination method.

[0047] The lower limb torque determination system, method, apparatus, computer equipment, and storage medium provided in this invention acquire a plantar pressure dataset, which contains time-series data. A multi-scale convolutional neural network is used to extract hierarchical features from the plantar pressure dataset, including local, intermediate, and global features. An attention-enhanced graph convolutional network is used to extract features from the plantar pressure dataset, obtaining different levels of time-series features. These time-series features are combined with attention to determine the weights of key nodes. A multi-source fusion method combining feature concatenation and weighted averaging is used to integrate the hierarchical features, time-series features, and key node weights into a unified feature vector. Residual connections preserve the original feature information, reducing information loss during feature processing, and the lower limb torque prediction result is calculated. This multi-stage data fusion algorithm, achieved through a multi-scale convolutional neural network and an attention-enhanced graph convolutional network, improves computational accuracy, accurately captures dynamic changes, and reveals the correlation between joint movements, thus enhancing the accuracy and efficiency of lower limb torque calculation. Attached Figure Description

[0048] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0049] Figure 1 This is a schematic diagram of the lower limb torque determination system of this application;

[0050] Figure 2 This is a schematic diagram of the arrangement of an IMU sensor and a foot pressure insole according to this application;

[0051] Figure 3This is a flowchart of one embodiment of the lower limb torque determination method of this application;

[0052] Figure 4 This is a flowchart illustrating one embodiment of the lower limb torque determination method of this application;

[0053] Figure 5 This is a schematic diagram of a structure of one embodiment of the lower limb torque determination device according to this application;

[0054] Figure 6 This is a schematic diagram of the structure of one embodiment of the computer device according to this application. Detailed Implementation

[0055] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains; the terminology used herein in the specification of the application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application; the terms "comprising" and "having," and any variations thereof, in the specification, claims, and foregoing drawings of this application, are intended to cover non-exclusive inclusion. The terms "first," "second," etc., in the specification, claims, or foregoing drawings of this application are used to distinguish different objects, not to describe a particular order.

[0056] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0057] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0058] Please see Figure 1 , Figure 1 This invention illustrates a lower limb torque determination system according to an embodiment of the present invention, comprising a sensor module 11, a data acquisition module 12, and a processing module 13, wherein the sensor module 11, the data acquisition module 12, and the processing module 13 are electrically connected; wherein,

[0059] The sensor module 11 acquires data through various sensors and transmits the acquired data to the data acquisition module. The sensor module 11 includes a plantar pressure controller 111 and an IMU sensor 112. The plantar pressure controller 111 is placed on the bottom of the wearable device or shoe to collect the contact force between the sole of the foot and the ground. The IMU sensor 112 is sparsely arranged and fixed on the joint of the lower limb to collect multidimensional acceleration and angular velocity data to reduce the complexity of sensor deployment and power consumption.

[0060] The data acquisition module 12 receives the data transmitted by the sensor module 11, performs standardized processing on the data, and then transmits it synchronously to the processing module. The data acquisition module 12 includes an STM32F105 microcontroller 121 and a Bluetooth device 122, which are used for data protocol conversion, high-precision time synchronization, adaptive data transmission, as well as data verification and automatic retransmission tasks.

[0061] The processing module 13 receives data from the data acquisition module 12 and determines the lower limb torque based on the received data.

[0062] like Figure 2 As shown, Figure 2 A schematic diagram of the sensor module 11 in this application is shown. A plantar pressure insole is placed inside a conventionally worn shoe, and a plantar pressure controller collects the contact force between the foot and the ground. An IMU sensor 112 is fixed to a joint area of ​​the human body to collect multidimensional acceleration and angular velocity data. The IMU sensor 112 is arranged in a sparse manner to reduce the complexity of hardware deployment and lower power consumption. This embodiment utilizes wearable sensors such as plantar pressure sensors and sparse IMUs for data acquisition, ensuring more accurate data collection.

[0063] Optionally, the Bluetooth device 122 employs cyclic redundancy check (CRC) to identify data transmission errors and, combined with a data retransmission strategy, improves data transmission reliability. The Bluetooth device 122 works in conjunction with the IMU sensor to achieve efficient data acquisition, clock synchronization, signal transmission management, and adaptive adjustment; error detection and automatic retransmission mechanisms ensure the stability and reliability of data transmission.

[0064] Optionally, the STM32F105 microcontroller 121 interacts with the plantar pressure controller 111 and the IMU sensor 112 to synchronously acquire data, and supports high-precision timestamp recording to ensure the consistency and integrity of the timing data.

[0065] After packaging the data, the data acquisition module 12 adds a timestamp and then transmits it synchronously via Bluetooth. A priority management mechanism ensures that critical data is transmitted first, reducing latency and guaranteeing data integrity. Simultaneously, it adaptively monitors Bluetooth transmission, automatically adjusting transmission power and channel to improve data transmission efficiency and stability, and reduce latency. A signal control and synchronization mechanism is also employed. The data acquisition module periodically sends synchronization signals to the processing module to ensure data consistency and real-time performance across modules. Error detection and automatic retransmission are also implemented. Bluetooth uses CRC checksum technology to determine data transmission errors; automatic retransmission is performed when errors occur, ensuring data reliability and transmission stability.

[0066] The processing module 13 employs a fusion algorithm combining multi-scale convolutional neural networks and long short-term memory networks to integrate multi-stage data and calculate joint torques. Specific steps include preliminary data correction and denoising, multi-scale feature extraction, time series modeling, fusion, and optimization. By utilizing multi-source data fusion strategies and Bayesian weight optimization mechanisms, the accuracy and response speed of data processing are further improved.

[0067] Please see Figure 3 , Figure 3 An embodiment of the present invention provides a method for determining lower limb torque, which is described in detail below:

[0068] S201: Obtain the plantar pressure dataset, which contains time-series data.

[0069] In one specific optional implementation, step S201, obtaining the plantar pressure dataset includes:

[0070] Receive real-time plantar pressure data, which includes plantar pressure, multidimensional acceleration, and angular velocity;

[0071] Add a timestamp to each real-time plantar pressure data point to obtain preliminary time-series data;

[0072] An adaptive Kalman filter algorithm was used to correct and denoise the initial time series data to obtain the plantar pressure dataset.

[0073] Specifically, the adaptive Kalman filter (ACTF) is an advanced filtering technique that can estimate and correct model and noise statistics using measurement data, thereby modifying the filter design and filtering error. This algorithm is particularly suitable for situations where the system and measurement noise characteristics are unknown or change over time. The core of the ACTF lies in dynamically adjusting the filter parameters, primarily the process noise covariance matrix (Q) and the measurement noise covariance matrix (R). These parameters are adjusted based on continuous evaluation of the innovation sequence (the difference between actual and expected values) to reduce estimation errors and anomaly detection thresholds.

[0074] In this embodiment, when processing plantar pressure datasets, the adaptive Kalman filter algorithm can be used for data correction and denoising. Specifically, the adaptive Kalman filter algorithm is implemented through the following steps:

[0075] State estimation: The algorithm first predicts the system state, which is usually based on the state and control input of the previous time step.

[0076] Innovation calculation: Innovation is the difference between the actual measurement and the measurement predicted based on the model. The adaptive Kalman filter algorithm calculates the innovation and adjusts the measurement noise variance according to the innovation covariance.

[0077] Outlier detection: Outliers in the measurement are detected by methods such as the chi-square test, and the measurement noise variance matrix is ​​adjusted when outliers are detected to suppress filter divergence.

[0078] Noise variance adjustment: Based on the ratio of the calculated to the predicted values ​​of the information covariance, the system noise variance matrix is ​​adjusted to improve filtering accuracy.

[0079] State update: The state estimate is updated using the adjusted Kalman gain and the new information.

[0080] Through these steps, the adaptive Kalman filter algorithm can effectively remove noise and correct the data from the time-series plantar pressure data, resulting in a more accurate and reliable dataset. This is crucial for subsequent data analysis and pattern recognition tasks, as accurate data can improve the performance and reliability of the model.

[0081] S202: Employs a multi-scale convolutional neural network to extract hierarchical features from the plantar pressure dataset. These hierarchical features include local, intermediate, and global features.

[0082] Multi-scale Convolutional Neural Networks (MSCNN) is a deep learning model that processes input data by extracting features at different scales. MSCNN extracts multi-scale features from the input data using convolutional kernels of varying sizes. These kernels at different scales can capture both details and global information in an image.

[0083] Local features, intermediate features, and global features can be defined at different levels according to the actual application needs.

[0084] In this embodiment, as Figure 4As shown, the multi-scale convolutional neural network (MSCNN) consists of three convolutional layers, incorporating ground pressure, multidimensional acceleration, and angular velocity. After initial processing with adaptive Kalman filtering to correct data, eliminate noise, and reduce errors, the data is then processed by the MSCNN to extract local, intermediate, and global features. The input data first passes through the first convolutional layer with a 3×3 kernel. Next, the ReLU activation function enhances the non-linear representation of features before entering the second convolutional layer with a 5×5 kernel. Simultaneously, batch normalization and the ReLU activation function are combined to improve feature extraction efficiency and convergence speed, enhancing the richness of feature representation. Finally, the data enters the third convolutional layer with a 7×7 kernel, passes through the ReLU activation function, and then through pooling and fully connected layers, ultimately outputting multi-scale features.

[0085] S203: A graph convolutional network based on attention mechanism is used to extract features from the plantar pressure dataset, obtain temporal features at different levels, and use the temporal features combined with attention to determine the weights of key nodes.

[0086] In one specific optional implementation, step S203, determining the key node weights by combining temporal features with attention, includes:

[0087] By employing an attention mechanism and temporal features, the contribution value of each node is calculated to obtain the feature importance weight value of each node.

[0088] Based on neighborhood relationships, the dynamic interaction characteristics between nodes are captured, and the correlation is calculated;

[0089] The weights of key nodes are determined based on the feature importance weights of each node and the correlations between nodes.

[0090] Specifically, time series modeling employs an attention-enhanced graph convolutional network. This network consists of three graph convolutional layers. Input data passes sequentially through the first (64 nodes), second (128 nodes), and third (256 nodes) convolutional layers. Each layer enhances nonlinear representation using the ReLU activation function, and batch normalization improves feature extraction performance and convergence speed. The attention mechanism calculates node feature importance weights based on node contribution values ​​to determine key time points; it then combines neighborhood information to capture the correlation between joints, thereby better capturing motion change characteristics.

[0091] In this embodiment, during the time series modeling stage, an attention-enhanced graph convolutional network (Attention-GCN) is used to extract time series features at different levels. The attention mechanism is combined to assign weights to the contributions of key nodes, thereby capturing the correlation between each joint.

[0092] S204: A multi-source fusion method combining feature splicing and weighted averaging is adopted to integrate hierarchical features, temporal features and key node weights into a unified feature vector; the original feature information is preserved through residual connection to reduce information loss in the feature processing process, and the lower limb torque prediction result is calculated.

[0093] In one specific optional implementation, a multi-source fusion method combining feature concatenation and weighted averaging is used to integrate hierarchical features, temporal features, and key node weights into a unified feature vector; after the step of retaining the original feature information through residual connection to reduce information loss during feature processing and calculating the lower limb torque prediction result, the lower limb torque determination method further includes:

[0094] The predicted lower limb torques were evaluated, and the evaluation results were obtained.

[0095] Based on the evaluation results, a Bayesian optimization mechanism is adopted to dynamically adjust the feature weight allocation strategy in order to further improve the model's predictive performance.

[0096] Specifically, after completing data fusion and outputting results, the next round of data acquisition and processing strategies are optimized by using feedback signals, reducing data transmission and processing volume, lowering energy consumption and latency, improving battery life, and meeting the requirements for long-term, uninterrupted, and continuous operation.

[0097] In this embodiment, a plantar pressure dataset is acquired, and the data in the plantar pressure dataset is time-series data. A multi-scale convolutional neural network is used to extract hierarchical features from the plantar pressure dataset, including local, intermediate, and global features. Based on an attention-enhanced graph convolutional network, features are extracted from the plantar pressure dataset to obtain time-series features at different levels. The weights of key nodes are determined by combining time-series features with attention. A multi-source fusion method combining feature concatenation and weighted averaging is used to integrate hierarchical features, time-series features, and key node weights into a unified feature vector. The original feature information is preserved through residual connections to reduce information loss during feature processing, and the lower limb torque prediction result is calculated. This multi-stage data fusion algorithm using a multi-scale convolutional neural network and an attention-enhanced graph convolutional network improves computational accuracy, accurately captures dynamic changes, and presents the correlation between joint movements, which is beneficial to improving the accuracy and efficiency of lower limb torque calculation.

[0098] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0099] Figure 5A schematic block diagram of a lower limb torque determination device corresponding to the lower limb torque determination method described in the above embodiments is shown. Figure 5 As shown, the lower limb torque determination device includes a data acquisition module 31, a first extraction module 32, a second extraction module 33, and a fusion calculation module 34. Detailed descriptions of each functional module are as follows:

[0100] Data acquisition module 31 is used to acquire plantar pressure dataset, in which the data is time series data;

[0101] The first extraction module 32 is used to extract hierarchical features from the plantar pressure dataset using a multi-scale convolutional neural network. The hierarchical features include local features, intermediate features, and global features.

[0102] The second extraction module 33 is used to extract features from the plantar pressure dataset using a graph convolutional network enhanced by an attention mechanism, to obtain temporal features at different levels, and to determine the weights of key nodes by combining the temporal features with attention.

[0103] The fusion calculation module 34 is used to integrate hierarchical features, temporal features and key node weights into a unified feature vector by using a multi-source fusion method that combines feature splicing and weighted averaging; the original feature information is preserved through residual connection to reduce information loss in the feature processing process, and the lower limb torque prediction result is calculated.

[0104] Optionally, the data acquisition module 31 includes:

[0105] The data receiving unit is used to receive real-time plantar pressure data, which includes plantar pressure, multidimensional acceleration, and angular velocity.

[0106] The timestamp addition unit is used to add a timestamp to each real-time plantar pressure data point to obtain preliminary time-series data.

[0107] The data preprocessing unit is used to perform data correction and noise reduction on the preliminary time series data using an adaptive Kalman filter algorithm to obtain the plantar pressure dataset.

[0108] Optionally, the second extraction module 33 includes:

[0109] The contribution calculation unit is used to calculate the contribution value of each node using an attention mechanism and temporal features, and obtain the feature importance weight value of each node.

[0110] The association computing unit is used to combine neighborhood information to capture the dynamic interaction characteristics between nodes;

[0111] The weight generation unit is used to determine the weight of key nodes based on the feature importance weight value of each node and the correlation between nodes.

[0112] Optionally, the lower limb torque determination device further includes:

[0113] The results evaluation module is used to evaluate the lower limb torque prediction results and obtain the evaluation results.

[0114] The weight update module is used to dynamically adjust the feature weight allocation strategy based on the evaluation results using a Bayesian optimization mechanism, in order to further improve the model's prediction performance.

[0115] Specific limitations regarding the lower limb torque determination device can be found in the limitations of the lower limb torque determination method described above, and will not be repeated here. Each module in the aforementioned lower limb torque determination device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the corresponding operations of each module.

[0116] To address the aforementioned technical problems, embodiments of this application also provide a computer device. Please refer to [link / reference needed]. Figure 6 , Figure 6 This is a basic structural block diagram of the computer device in this embodiment.

[0117] The computer device 4 includes a memory 41, a processor 42, and a network interface 43 that are interconnected via a system bus. It should be noted that only the computer device 4 with components connected to the memory 41, processor 42, and network interface 43 is shown in the figure; however, it should be understood that it is not required to implement all the components shown, and more or fewer components can be implemented alternatively. Those skilled in the art will understand that the computer device described here is a device capable of automatically performing numerical calculations and / or information processing according to pre-set or stored instructions, and its hardware includes, but is not limited to, microprocessors, application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), digital signal processors (DSPs), embedded devices, etc.

[0118] The computer device can be a desktop computer, laptop, handheld computer, or cloud server, etc. The computer device can interact with the user via a keyboard, mouse, remote control, touchpad, or voice control.

[0119] The memory 41 includes at least one type of readable storage medium, including flash memory, hard disk, multimedia card, card-type memory (e.g., SD or D-interface display memory), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, disk, optical disk, etc. In some embodiments, the memory 41 may be an internal storage unit of the computer device 4, such as the hard disk or memory of the computer device 4. In other embodiments, the memory 41 may also be an external storage device of the computer device 4, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc., equipped on the computer device 4. Of course, the memory 41 may include both the internal storage unit and its external storage device of the computer device 4. In this embodiment, the memory 41 is typically used to store the operating system and various application software installed on the computer device 4, such as the program code of the lower limb torque determination method. In addition, the memory 41 can also be used to temporarily store various types of data that have been output or will be output.

[0120] In some embodiments, the processor 42 may be a central processing unit (CPU), controller, microcontroller, microprocessor, or other data processing chip. The processor 42 is typically used to control the overall operation of the computer device 4. In this embodiment, the processor 42 is used to run program code stored in the memory 41 or process data, for example, to run program code for a lower limb torque determination method.

[0121] The network interface 43 may include a wireless network interface or a wired network interface, which is typically used to establish communication connections between the computer device 4 and other electronic devices.

[0122] This application also provides another embodiment, namely, a computer-readable storage medium storing an interface display program that can be executed by at least one processor to cause the at least one processor to perform the steps of the lower limb torque determination method as described above.

[0123] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0124] Obviously, the embodiments described above are only some embodiments of this application, not all embodiments. The accompanying drawings show preferred embodiments of this application, but do not limit the patent scope of this application. This application can be implemented in many different forms; rather, the purpose of providing these embodiments is to provide a more thorough and comprehensive understanding of the disclosure of this application. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing specific embodiments, or make equivalent substitutions for some of the technical features. Any equivalent structures made using the content of this application's specification and drawings, directly or indirectly applied to other related technical fields, are similarly within the scope of patent protection of this application.

Claims

1. A method for determining lower limb torque, characterized in that, An application is made to a lower limb torque determination system, which includes a sensor module, a data acquisition module, and a processing module, wherein the sensor module, the data acquisition module, and the processing module are electrically connected. The sensor module acquires data through various sensors and transmits the acquired data to the data acquisition module. The sensor module includes a plantar pressure controller and an IMU sensor. The plantar pressure controller is placed on the bottom of the wearable device or shoe to collect the contact force between the sole of the foot and the ground. The IMU sensor is sparsely arranged and fixed to the joint of the lower limb to collect multidimensional acceleration and angular velocity data. The data acquisition module receives data transmitted by the sensor module, performs standardized processing on the data, and then synchronously transmits it to the processing module. The data acquisition module includes an STM32F105 microcontroller and a Bluetooth device, which are used for data protocol conversion, high-precision time synchronization, adaptive data transmission, data verification, and automatic retransmission tasks. The STM32F105 microcontroller interacts with the plantar pressure controller and the IMU sensor to synchronously acquire data and supports high-precision timestamp recording. The processing module receives data from the data acquisition module and determines the lower limb torque based on the received data; The method for determining the lower limb torque includes: Acquiring a plantar pressure dataset, wherein the data in the plantar pressure dataset is time-series data; the acquisition of the plantar pressure dataset includes: receiving real-time plantar pressure data, wherein the real-time plantar pressure data includes plantar pressure, multidimensional acceleration, and angular velocity; adding a timestamp to each real-time plantar pressure data to obtain preliminary time-series data; and using an adaptive Kalman filter algorithm to perform data correction and noise reduction on the preliminary time-series data to obtain the plantar pressure dataset. A multi-scale convolutional neural network is used to extract hierarchical features from the plantar pressure dataset. The hierarchical features include local features, intermediate features, and global features, which reflect the dynamic changes of plantar pressure at different levels. Based on an attention-enhanced graph convolutional network, features are extracted from the plantar pressure dataset to obtain temporal features at different levels. The temporal features are then combined with attention to determine the weights of key nodes. A multi-source fusion and residual connection method combining feature splicing and weighted averaging is adopted to integrate hierarchical features, temporal features and key node weights into a unified feature vector; then, the original feature information is preserved through residual connection to obtain the lower limb torque prediction result.

2. The method for determining lower limb torque as described in claim 1, characterized in that, The step of determining the key node weights by combining the temporal features with attention includes: Using an attention mechanism and the aforementioned temporal features, the contribution value of each node is calculated to obtain the feature importance weight value of each node. Based on neighborhood relationships, the dynamic interaction characteristics between nodes are captured, and the correlation is calculated; The weights of key nodes are determined based on the feature importance weights of each node and the correlations between nodes.

3. The method for determining lower limb torque as described in claim 1, characterized in that, A multi-source fusion method combining feature concatenation and weighted averaging is adopted to integrate hierarchical features, temporal features, and key node weights into a unified feature vector; After the step of preserving the original feature information through residual connection and calculating the lower limb torque prediction result, the lower limb torque determination method further includes: The predicted lower limb torque is evaluated to obtain the evaluation result; Based on the evaluation results, a Bayesian optimization mechanism is used to dynamically adjust the feature weight allocation strategy and dynamically optimize the feature weights.

4. A device for determining lower limb torque, characterized in that, The lower limb torque determination method according to any one of claims 1 to 3, wherein the lower limb torque determination device comprises: The data acquisition module is used to acquire plantar pressure dataset, wherein the data in the plantar pressure dataset is time-series data; The first extraction module is used to extract hierarchical features from the plantar pressure dataset using a multi-scale convolutional neural network. The hierarchical features include local features, intermediate features, and global features, which reflect the dynamic changes of plantar pressure at different levels. The second extraction module is used to extract features from the plantar pressure dataset based on an attention-enhanced graph convolutional network to obtain temporal features at different levels, and to use the temporal features in combination with attention to determine the weights of key nodes. The fusion calculation module is used to integrate hierarchical features, temporal features, and key node weights into a unified feature vector by using a multi-source fusion method that combines feature splicing and weighted averaging. The original feature information is preserved through residual connection, and the lower limb torque prediction result is calculated.

5. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the lower limb torque determination method as described in any one of claims 1 to 3.

6. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the lower limb torque determination method as described in any one of claims 1 to 3.

Images