Cloud-platform-supported full-period management method for health of foot spines of teenagers

Abstract

The invention relates to the technical field of health management. The teenager foot spine health full-period management method supported by a cloud platform comprises the following steps: acquiring plantar pressure distribution data, spine posture data and user behavior data of a user in real time; based on the joint feature vector, dynamic training parameter optimization processing is carried out through a reinforcement learning algorithm, and a dynamic training parameter combination is obtained; based on the progress speed of the user, carrying out training period adjustment processing to obtain an adaptive training period; performing new action generation processing by utilizing a generative adversarial network to obtain variation training actions, and performing similarity screening processing on the variation training actions to obtain a low-similarity action set; on the basis of the low-similarity action set, the dynamic training parameter combination and 3D printing orthopedic appliance parameters are subjected to association processing, and a personalized foot spine health management scheme is generated through a preset cloud platform, so that the problems that a static monitoring means is single, a training scheme is fixed, and adjustment lags behind growth and development change are solved.

Description

Technical Field

[0001] The present invention relates to the field of health management technology, and in particular to a full-cycle management method for adolescent foot spine health supported by a cloud platform. Background Art

[0002] During the growth of adolescents, the health of the foot spine is crucial to physical development. The normal structure of the foot can buffer the impact of movement and protect the bone joints. At the same time, the stability and functional state of the foot will have a direct impact on the spine. Abnormalities in the foot may change the force of the spine and lead to abnormal spinal posture. Healthy spinal posture helps maintain body balance and coordination and promote the normal development of internal organs. If there is a problem with the foot spine, it may cause abnormal posture and pain diseases, affecting healthy growth.

[0003] However, traditional foot and spine health management has the problem of single static detection method leading to missing dynamic activity data and causing risk of delayed intervention; at the same time, there is the problem of fixed training programs and adjustments lagging behind growth and development changes, resulting in attenuated correction effects. Summary of the invention

[0004] Based on this, it is necessary to provide a full-cycle management method for adolescent foot and spine health supported by a cloud platform to address the above-mentioned technical issues, so as to solve the problems of single static monitoring methods, rigid training programs and adjustments lagging behind growth and development changes.

[0005] This application provides a full-cycle management method for adolescent foot and spine health supported by a cloud platform, including:

[0006] Obtain the user's plantar pressure distribution data, spinal posture data and user behavior data in real time. The user behavior data includes the user's daily steps, exercise duration and training compliance score;

[0007] The plantar pressure distribution data is processed by temporal and spatial feature extraction to obtain the foot pressure feature vector; the spinal posture data is processed by stability index calculation to obtain the spinal stability index; the foot pressure feature vector, spinal stability index and user behavior data are fused to obtain the joint feature vector;

[0008] Based on the joint feature vector, the dynamic training parameters are optimized by the reinforcement learning algorithm to obtain the dynamic training parameter combination, which includes training intensity, complexity and action variants;

[0009] Based on the user's progress speed, the training cycle is adjusted through the proportional-integral-derivative control algorithm to obtain an adaptive training cycle;

[0010] Based on the joint feature vector and the adaptive training cycle, use the generative adversarial network to perform new action generation processing to obtain mutated training actions, and perform similarity screening processing on the mutated training actions to obtain a set of low-similarity actions;

[0011] Based on the set of low-similarity actions, associate the dynamic training parameter combination with the 3D printing orthosis parameters, and generate a personalized foot and spine health management plan through a preset cloud platform. The personalized foot and spine health management plan includes a dynamically updated training plan, orthosis adaptation parameters, and a long-term tracking strategy.

[0012] Furthermore, based on the joint feature vector, perform dynamic training parameter optimization processing through a reinforcement learning algorithm to obtain a dynamic training parameter combination, including:

[0013] Based on the joint feature vector, perform reinforcement learning state space and action space definition processing to obtain a state space and an action space;

[0014] Based on the state space and the action space, use the following formula to perform reward function calculation processing to obtain a reward value:

[0015] Reward value = 0.6×ΔP target - 0.3×ΔI s + 0.1×C

[0016] Where, ΔP target represents the pressure improvement rate of the plantar pressure feature vector in the target area, ΔI s represents the absolute value of the change in the spinal stability index, and C represents the training compliance score;

[0017] Based on the reward value, perform policy network parameter update processing through the deep deterministic policy gradient algorithm to obtain a dynamic training parameter combination.

[0018] Furthermore, based on the user's progress speed, perform training cycle adjustment processing through a proportional-integral-derivative control algorithm to obtain an adaptive training cycle, including:

[0019] Use the following formula to calculate the user's progress speed based on the user behavior data:

[0020]

[0021] Where, V represents the user's learning progress speed, m represents the number of learning behavior records, L i represents the i-th learning evaluation score, α represents the progress speed coefficient, β i represents the weight factor of the i-th learning, R represents the user's comprehensive progress speed, p represents the number of evaluation indicators, E k represents the score of the k-th evaluation indicator, Wk denotes the weight of the k-th evaluation index;

[0022] Based on the progress speed, the training cycle is adjusted through a proportional-integral-derivative control algorithm to obtain an adaptive training cycle.

[0023] Furthermore, based on the joint feature vector and the adaptive training cycle, a generative adversarial network is used to generate new actions, resulting in mutated training actions. Then, a similarity screening process is performed on the mutated training actions to obtain a set of low-similarity actions, including:

[0024] Based on the joint feature vector and the adaptive training cycle, new actions are generated through a generative adversarial network to obtain mutated training actions;

[0025] Based on the mutated training actions and the historical action parameter vector, a cosine similarity calculation process is performed to obtain a similarity value;

[0026] Based on the similarity value, a threshold screening process is performed to obtain a set of low-similarity actions.

[0027] Furthermore, spatio-temporal feature extraction is performed on the plantar pressure distribution data to obtain a plantar pressure feature vector; a stability index calculation is performed on the spinal posture data to obtain a spinal stability index; the plantar pressure feature vector, the spinal stability index, and the user behavior data are fused to obtain a joint feature vector, including:

[0028] The following formula is used to perform principal component analysis on the plantar pressure distribution data to obtain a plantar pressure feature vector:

[0029]

[0030] λν = Cν

[0031] Y = W T (X - μ)

[0032] where C represents the covariance matrix, n represents the number of samples, x i represents the data vector of the i-th plantar pressure sampling point, μ represents the mean vector of all sampling points, λ represents the eigenvalue, ν represents the eigenvector, Y represents the feature vector after dimensionality reduction, W represents the eigenvector matrix, and X represents the original plantar pressure data matrix;

[0033] A stability index calculation is performed on the spinal posture data to obtain a spinal stability index;

[0034] The plantar pressure feature vector, the spinal stability index, and the user behavior data are input into a trained autoencoder neural network for feature fusion to obtain a joint feature vector.

[0035] Further, based on the low-similarity action set, the dynamic training parameter combination is associated with the 3D printing orthosis parameters, and a personalized foot and spine health management plan is generated through a preset cloud platform, including:

[0036] Based on the low-similarity action set and the dynamic training parameters, the spatio-temporal features of the actions are extracted through the principal component analysis algorithm to generate action feature vectors;

[0037] The Euclidean distance matching process is performed on the action feature vectors and the preset 3D printing orthosis parameter database to obtain the target appliance parameter combination;

[0038] Based on the target appliance parameter combination, a modeling process is performed through the multiple linear regression algorithm to obtain a personalized foot and spine health management model;

[0039] According to the personalized foot and spine health management model, the correlation coefficient calculation process is performed on the dynamic training parameters and the 3D printing orthosis parameters to obtain a parameter correlation matrix;

[0040] Based on the parameter correlation matrix, the optimal matching degree calculation process is performed through the gradient descent algorithm to obtain the adjusted orthosis parameters;

[0041] The adjusted orthosis parameters and the dynamic training parameters are fused to generate a personalized foot and spine health management plan.

[0042] The technical solution provided by this application includes the following technical effects: By providing a full-cycle management method for adolescent foot and spine health supported by a cloud platform, including: real-time acquisition of the user's plantar pressure distribution data, spinal posture data, and user behavior data, where the user behavior data includes the user's daily steps, exercise duration, and training compliance score; performing spatio-temporal feature extraction processing on the plantar pressure distribution data to obtain a plantar pressure feature vector; performing stability index calculation processing on the spinal posture data to obtain a spinal stability index; fusing the plantar pressure feature vector, spinal stability index, and user behavior data to obtain a joint feature vector; based on the joint feature vector, performing dynamic training parameter optimization processing through a reinforcement learning algorithm to obtain a dynamic training parameter combination, where the dynamic training parameter combination includes training intensity, complexity, and action variants; based on the user's progress speed, performing training cycle adjustment processing through a proportional-integral-derivative control algorithm to obtain an adaptive training cycle; based on the joint feature vector and the adaptive training cycle, using a generative adversarial network to perform new action generation processing to obtain mutated training actions, and performing similarity screening processing on the mutated training actions to obtain a set of low-similarity actions; based on the set of low-similarity actions, associating the dynamic training parameter combination with the 3D printing orthosis parameters, and generating a personalized foot and spine health management plan through a preset cloud platform, where the personalized foot and spine health management plan includes a dynamically updated training plan, orthosis adaptation parameters, and a long-term tracking strategy, to solve the problems of single static monitoring means, fixed training plans, and lagging adjustment behind growth and development changes. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the following will briefly introduce the drawings required for use in the description of the embodiments or related technologies. Obviously, the drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0044] Figure 1 It is a flowchart of the full-cycle management method for adolescent foot and spine health supported by a cloud platform in an embodiment of the present invention;

[0045] Figure 2 It is a flowchart of the full-cycle management method for adolescent foot and spine health supported by a cloud platform in an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0046] In order to make the objectives, technical solutions, and advantages of the present application clearer, the following further details the present application in conjunction with the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0047] AsFigure 1 As shown in and

[0048] , the present application provides a full-cycle management method for adolescent foot and spine health supported by a cloud platform, including:

[0048] S101: Real-time obtain the plantar pressure distribution data, spinal posture data and user behavior data of the user. The user behavior data includes the user's daily steps, exercise duration and training compliance score.

[0049] Specifically, the pressure sensors embedded in the insole are used to capture the pressure data of each area of the sole in real time, and are transmitted wirelessly to a mobile device or a cloud platform. After denoising and signal processing, accurate pressure distribution information is obtained. A 3D spine assessment system or a wearable posture monitoring device is used to capture the shape and posture parameters of the spine, such as the bending angle, inclination degree, etc., and generate a detailed three-dimensional model or data record to evaluate the health status of the spine. Wearable devices such as smart bracelets and smart watches are used to automatically record data such as the user's daily steps and exercise duration, and information such as the user's training compliance score is obtained through a mobile application. These data will be combined with the plantar pressure and spinal posture data to provide support for comprehensively evaluating the foot and spine health status of the user.

[0050] In one embodiment, multiple cameras are used to take pictures of the user's feet and spine from different angles to ensure comprehensive morphological information is obtained. Using three-dimensional reconstruction technology, the multi-angle pictures are transformed into accurate three-dimensional models of the feet and spine for detailed analysis. Through a plantar pressure tester, the plantar pressure distribution data of the user in a static and dynamic state is obtained. A three-dimensional motion capture system (such as BTS POSEIDON) or a wearable posture monitoring device is used to capture the shape and motion data of the spine. The above data is integrated to provide a comprehensive basis for formulating a personalized foot and spine health management plan. In addition, the user can take multi-angle photos of the feet and spine through a mobile terminal, and after being processed by three-dimensional reconstruction technology, three-dimensional images of the feet and spine are generated to achieve multi-dimensional imaging.

[0051] S102: Perform spatio-temporal feature extraction processing on the plantar pressure distribution data to obtain a plantar pressure feature vector; perform stability index calculation processing on the spinal posture data to obtain a spinal stability index; fuse the plantar pressure feature vector, spinal stability index and user behavior data to obtain a joint feature vector.

[0052] Specifically, plantar pressure data is obtained through sensors. After preprocessing, spatio-temporal features such as pressure change rate and distribution range are extracted to form a plantar pressure feature vector. Spinal posture data is obtained using a 3D evaluation system or wearable device, and a stability index is calculated. Quantification indicators are obtained by analyzing the volatility of posture parameters. Data fusion: After standardizing the plantar pressure feature vector, spinal stability index, and user behavior data (such as the number of steps and exercise duration), they are integrated into a joint feature vector using feature splicing or weighted fusion methods to provide data support for subsequent analysis.

[0053] S103: Based on the joint feature vector, dynamic training parameter optimization is performed through a reinforcement learning algorithm to obtain a dynamic training parameter combination, which includes training intensity, complexity, and action variants.

[0054] Specifically, first, the joint feature vector is used as the state representation in the reinforcement learning algorithm, and the state space and action space of the reinforcement learning are defined. The state space contains information in multiple aspects such as plantar pressure distribution, spinal posture, and user behavior, and the action space includes a dynamic training parameter combination in dimensions such as training intensity, complexity, and action variants.

[0055] Next, based on the state space and action space, a reward value is obtained through calculation using a reward function. The reward function comprehensively considers multiple factors such as the pressure improvement rate of the plantar pressure feature vector in the target area, the absolute value of the change in the spinal stability index, and the training compliance score to comprehensively measure the training effect and user progress.

[0056] Finally, based on the reward value, the policy network parameter update is performed through the Deep Deterministic Policy Gradient algorithm (DDPG) to obtain a dynamic training parameter combination. The DDPG algorithm combines deterministic policies and deep neural networks and is suitable for problems with continuous action spaces. It generates actions (i.e., dynamic training parameter combinations) through the Actor network, evaluates the quality of actions through the Critic network, and uses an experience replay mechanism and a target network to stabilize the training process. Ultimately, the optimal training parameter combination for the user's current state is optimized to improve the effect and efficiency of foot and spine health training.

[0057] S104: Based on the user's progress speed, the training cycle is adjusted through a proportional-integral-derivative control algorithm to obtain an adaptive training cycle.

[0058] Specifically, first, the user's progress speed is input as a feedback signal into the PID controller. Here, the progress speed can be measured by the improvement of various indicators during the user's training, such as the optimization of plantar pressure distribution, the improvement of spinal posture, and the improvement of training performance.

[0059] Then, the PID controller calculates according to the deviation between the progress speed and the expected goal, following the three control laws of proportional (P), integral (I), and derivative (D). Proportional control quickly adjusts the training cycle according to the magnitude and direction of the current deviation to reduce the deviation; integral control considers the accumulated deviation in the past to eliminate long-term steady-state errors and ensure that the adjustment of the training cycle can continuously and effectively promote the user's progress; derivative control predicts the change trend of the deviation and adjusts the training cycle in advance to avoid over-adjustment and oscillation.

[0060] S105: Based on the joint feature vector and the adaptive training cycle, use a generative adversarial network to perform new action generation processing to obtain mutant training actions, and perform similarity screening processing on the mutant training actions to obtain a set of low-similarity actions.

[0061] Specifically, first, use the joint feature vector and the adaptive training cycle as the input of the generative adversarial network (GAN). Among them, the joint feature vector contains information in multiple aspects such as plantar pressure distribution, spinal posture, and user behavior, and the adaptive training cycle is dynamically adjusted according to the user's progress speed to ensure the pertinence and effectiveness of training. Then, the generator in the GAN generates a series of new training actions according to the input joint feature vector and adaptive training cycle. These new actions are designed to increase the diversity and pertinence of training and avoid the adaptive plateau effect caused by the user repeating the same actions. By learning the distribution of real data, the generator can generate new actions that are highly similar to real training actions but have variability.

[0062] Next, perform similarity screening processing on the generated mutant training actions. This step screens out a set of low-similarity actions by calculating the similarity between the newly generated actions and the existing actions. The calculation of similarity can be based on multiple dimensions of the actions, such as the intensity, complexity, and involved muscle groups of the actions, to ensure that the screened actions have sufficient differences in multiple aspects, thereby providing the user with a fresh and challenging training experience and further improving the training effect. Through the above process, personalized training actions can be dynamically provided for the user to meet their needs in different training cycles and promote the overall improvement of foot and spine health.

[0063] S106: Based on the set of low-similarity actions, associate the dynamic training parameter combination with the 3D printing orthosis parameters, and generate a personalized foot and spine health management plan through a pre-set cloud platform. The personalized foot and spine health management plan includes a dynamically updated training plan, orthosis adaptation parameters, and a long-term tracking strategy.

[0064] Specifically, first, combine the low-similarity action set with the dynamic training parameter combination to form a diverse training action library. Each action is labeled according to its intensity, complexity, and variants to match the needs of different users at different training stages. Then, associate these training actions with the 3D printing orthosis parameters. The parameters of the 3D printing orthosis, such as the curvature of the insole and the support position, are customized according to the user's plantar pressure distribution and spinal posture data to provide the best biomechanical support. Then, through a pre-set cloud platform, integrate the above data and generate a personalized foot and spine health management plan. The cloud platform uses its powerful data processing and storage capabilities, combined with the user's real-time data, to dynamically adjust the training plan and orthosis parameters to ensure the pertinence and effectiveness of the plan.

[0065] Finally, the personalized foot and spine health management plan includes a dynamically updated training plan, orthosis adaptation parameters, and a long-term tracking strategy. The training plan is adjusted according to the user's progress speed and adaptive training cycle, the orthosis parameters are optimized according to the changes in the user's foot and spine state, and the long-term tracking strategy continuously monitors the user's health status to ensure the long-term effectiveness and stability of the management plan.

[0066] The technical solution provided by this application includes the following technical effects: By providing a cloud platform-supported full-cycle management method for adolescent foot and spine health, including: real-time obtaining the user's plantar pressure distribution data, spinal posture data, and user behavior data, where the user behavior data includes the user's daily steps, exercise duration, and training compliance score; performing spatio-temporal feature extraction processing on the plantar pressure distribution data to obtain a plantar pressure feature vector; performing stability index calculation processing on the spinal posture data to obtain a spinal stability index; fusing the plantar pressure feature vector, spinal stability index, and user behavior data to obtain a joint feature vector; based on the joint feature vector, performing dynamic training parameter optimization processing through a reinforcement learning algorithm to obtain a dynamic training parameter combination, where the dynamic training parameter combination includes training intensity, complexity, and action variants; based on the user's progress speed, performing training cycle adjustment processing through a proportional-integral-derivative control algorithm to obtain an adaptive training cycle; based on the joint feature vector and the adaptive training cycle, using a generative adversarial network to perform new action generation processing to obtain mutated training actions, and performing similarity screening processing on the mutated training actions to obtain a low-similarity action set; based on the low-similarity action set, associating the dynamic training parameter combination with the 3D printing orthosis parameters, and generating a personalized foot and spine health management plan through a pre-set cloud platform, where the personalized foot and spine health management plan includes a dynamically updated training plan, orthosis adaptation parameters, and a long-term tracking strategy, to solve the problems of single static monitoring means, fixed training plans, and lagging adjustment behind growth and development changes.

[0067] Furthermore, based on the joint feature vector, dynamic training parameter optimization is performed through a reinforcement learning algorithm to obtain a dynamic training parameter combination, including:

[0068] Based on the joint feature vector, define the state space and action space of reinforcement learning to obtain the state space and action space;

[0069] Based on the state space and action space, use the following formula to calculate the reward function and obtain the reward value:

[0070] Reward value = 0.6×ΔP target - 0.3×ΔI s + 0.1×C

[0071] Where, ΔP target represents the pressure improvement rate of the plantar pressure feature vector in the target area, and ΔI s represents the absolute value of the change in the spinal stability index, and C represents the training compliance score;

[0072] Based on the reward value, update the parameters of the policy network through the Deep Deterministic Policy Gradient algorithm to obtain the dynamic training parameter combination.

[0073] Specifically, first, use the joint feature vector as the state representation in the reinforcement learning algorithm, and define the state space and action space of reinforcement learning. The state space contains information such as plantar pressure distribution, spinal posture, and user behavior, and the action space includes a combination of dynamic training parameters such as training intensity, complexity, and action variants.

[0074] Next, based on the state space and action space, calculate the reward function to obtain the reward value. The reward function comprehensively considers multiple factors such as the pressure improvement rate of the plantar pressure feature vector in the target area, the absolute value of the change in the spinal stability index, and the training compliance score to comprehensively measure the training effect and user progress.

[0075] Finally, based on the reward value, update the parameters of the policy network through the Deep Deterministic Policy Gradient (DDPG) algorithm to obtain the dynamic training parameter combination. The DDPG algorithm combines deep neural networks and deterministic policy gradients and is suitable for problems with continuous action spaces. It generates actions (i.e., dynamic training parameter combinations) through the Actor network, evaluates the quality of the actions through the Critic network, uses the experience replay mechanism and the target network to stabilize the training process, and finally optimizes the optimal training parameter combination for the user's current state to improve the effect and efficiency of foot and spine health training.

[0076] Furthermore, based on the user's progress speed, adjust the training cycle through the proportional-integral-derivative control algorithm to obtain an adaptive training cycle, including:

[0077] Using the following formula, based on user behavior data, calculate the user's progress speed:

[0078]

[0079] Where V represents the user's learning progress speed, m represents the number of learning behavior records, L i represents the score of the i-th learning assessment, α represents the progress speed coefficient, β i represents the weight factor of the i-th learning, R represents the user's comprehensive progress speed, p represents the number of evaluation indicators, E k represents the score of the k-th evaluation indicator, W k represents the weight of the k-th evaluation indicator;

[0080] Based on the progress speed, perform training cycle adjustment processing through a proportional-integral-derivative control algorithm to obtain an adaptive training cycle.

[0081] Specifically, first, use the joint feature vector as the state representation in the reinforcement learning algorithm, and define the state space and action space of the reinforcement learning. The state space contains information in multiple aspects such as plantar pressure distribution, spinal posture, and user behavior, and the action space includes a combination of dynamic training parameters in dimensions such as training intensity, complexity, and action variants.

[0082] Next, based on the state space and action space, perform calculation processing using the reward function to obtain a reward value. The reward function comprehensively considers multiple factors such as the pressure improvement rate of the plantar pressure feature vector in the target area, the absolute value of the change in the spinal stability index, and the training compliance score, etc., to comprehensively measure the training effect and the user's progress.

[0083] Finally, based on the reward value, perform policy network parameter update processing through the Deep Deterministic Policy Gradient algorithm (DDPG) to obtain a combination of dynamic training parameters. The DDPG algorithm combines a deterministic policy and a deep neural network, and is suitable for problems with a continuous action space. It generates actions (i.e., a combination of dynamic training parameters) through the Actor network, the Critic network evaluates the quality of the actions, and uses an experience replay mechanism and a target network to stabilize the training process, and finally optimizes the optimal training parameter combination for the user's current state to improve the effect and efficiency of the foot and spine health training.

[0084] Furthermore, based on the joint feature vector and the adaptive training cycle, use a generative adversarial network to perform new action generation processing to obtain mutant training actions, and perform similarity screening processing on the mutant training actions to obtain a set of low-similarity actions, including:

[0085] New action generation processing is performed through a generative adversarial network based on a combined feature vector and an adaptive training cycle to obtain mutated training actions;

[0086] Cosine similarity calculation processing is performed based on the mutated training actions and the historical action parameter vector to obtain a similarity value;

[0087] Threshold screening processing is performed based on the similarity value to obtain a collection of low-similarity actions.

[0088] Specifically, first, the combined feature vector and the adaptive training cycle are used as inputs to a generative adversarial network (GAN). The combined feature vector contains information from multiple aspects such as plantar pressure distribution, spinal posture, and user behavior. The adaptive training cycle is dynamically adjusted according to the user's progress speed to ensure the pertinence and effectiveness of training. The generator in the GAN generates a series of new training actions based on these inputs. These actions are designed to increase the diversity and pertinence of training and prevent the user from experiencing an adaptive plateau effect due to repeating the same actions.

[0089] Next, cosine similarity calculation processing is performed on the generated mutated training actions and the historical action parameter vector to obtain a similarity value. Cosine similarity is used to measure the similarity between two vectors. By calculating the cosine value of the mutated action and the historical action in multiple dimensions (such as the intensity, complexity, and involved muscle groups of the action), their similarity is obtained.

[0090] Finally, threshold screening processing is performed based on the similarity value to obtain a set of low-similarity actions. A similarity threshold is set, and actions with a similarity lower than this threshold are screened out to form a set of low-similarity actions. These actions have sufficient differences from the historical actions in multiple aspects, thus providing the user with a fresh and challenging training experience and further improving the training effect.

[0091] Furthermore, spatio-temporal feature extraction processing is performed on the plantar pressure distribution data to obtain a plantar pressure feature vector; stability index calculation processing is performed on the spinal posture data to obtain a spinal stability index; the plantar pressure feature vector, the spinal stability index, and the user behavior data are fused to obtain a combined feature vector, including:

[0092] The following formula is used to perform principal component analysis processing on the plantar pressure distribution data to obtain a plantar pressure feature vector:

[0093]

[0094] λν = Cν

[0095] Y = W T (X - μ)

[0096] Among them, C represents the covariance matrix, n represents the number of samples, and x i represents the data vector of the i-th plantar pressure sampling point, μ represents the mean vector of all sampling points, λ represents the eigenvalue, ν represents the eigenvector, Y represents the eigenvector after dimensionality reduction, W represents the eigenvector matrix, and X represents the original plantar pressure data matrix;

[0097] Perform stability index calculation processing on the spinal posture data to obtain the spinal stability index;

[0098] Input the plantar pressure eigenvector, spinal stability index, and user behavior data into the trained autoencoder neural network for feature fusion processing to obtain the joint eigenvector.

[0099] Specifically, first, use principal component analysis (PCA) to process the plantar pressure distribution data to obtain the plantar pressure eigenvector. The specific steps are as follows: calculate the covariance matrix, solve the eigenvalues and eigenvectors, select the main eigenvectors for dimensionality reduction, and obtain the eigenvector after dimensionality reduction, which is the plantar pressure eigenvector.

[0100] Secondly, perform stability index calculation processing on the spinal posture data to obtain the spinal stability index. The specific method is determined according to the characteristics of the spinal posture data.

[0101] Finally, input the plantar pressure eigenvector, spinal stability index, and user behavior data into the trained autoencoder neural network for feature fusion processing to obtain the joint eigenvector. The autoencoder neural network compresses the input data into a low-dimensional feature representation through the encoder and then reconstructs the input data through the decoder, thereby realizing the fusion and extraction of features.

[0102] Furthermore, as Figure 2 shown, based on the low-similarity action set, associate the dynamic training parameter combination with the 3D printing orthosis parameters, and generate a personalized foot and spine health management plan through the preset cloud platform, including:

[0103] S201: Based on the low-similarity action set and dynamic training parameters, perform action spatio-temporal feature extraction processing through the principal component analysis algorithm to generate an action feature vector;

[0104] S202: Perform Euclidean distance matching processing on the action feature vector and the preset 3D printing orthosis parameter database to obtain the target appliance parameter combination;

[0105] S203: Based on the target appliance parameter combination, perform modeling processing through the multiple linear regression algorithm to obtain a personalized foot and spine health management model;

[0106] S204: Calculate the correlation coefficient between the dynamic training parameters and the 3D printing orthosis parameters according to the personalized foot and spinal health management model to obtain a parameter correlation matrix;

[0107] S205: Based on the parameter correlation matrix, perform optimal matching degree calculation through the gradient descent algorithm to obtain the adjusted orthosis parameters;

[0108] S206: Integrate the adjusted orthosis parameters and the dynamic training parameters to generate a personalized foot and spinal health management plan.

[0109] Specifically, first, based on the low-similarity action set and the dynamic training parameters, perform action spatio-temporal feature extraction through the principal component analysis algorithm to generate action feature vectors. Among them, the principal component analysis algorithm can reduce the dimension of high-dimensional action data, extract the principal components that best represent the action features, and form action feature vectors.

[0110] Next, perform Euclidean distance matching on the action feature vectors and the preset 3D printing orthosis parameter database to obtain the target appliance parameter combination. As a commonly used geometric distance measurement method, the Euclidean distance can effectively measure the similarity between the action feature vectors and different orthosis parameters, so as to find the optimal matching appliance parameter combination.

[0111] Then, based on the target appliance parameter combination, perform modeling through the multiple linear regression algorithm to obtain a personalized foot and spinal health management model. The multiple linear regression algorithm can analyze the linear relationship between multiple independent variables and one dependent variable, and help establish an association model between the action features and the orthosis parameters.

[0112] Next, according to the personalized foot and spinal health management model, calculate the correlation coefficient between the dynamic training parameters and the 3D printing orthosis parameters to obtain a parameter correlation matrix. This step forms a correlation matrix by calculating the correlation between different parameters, providing a basis for subsequent parameter optimization. Then, based on the parameter correlation matrix, perform optimal matching degree calculation through the gradient descent algorithm to obtain the adjusted orthosis parameters. As an optimization algorithm, the gradient descent algorithm can update the parameters iteratively, minimize the error of the model, and find the optimal parameter combination.

[0113] Finally, integrate the adjusted orthosis parameters and the dynamic training parameters to generate a personalized foot and spinal health management plan. This step integrates the optimized parameters into the cloud platform to form a comprehensive and personalized health management plan, providing the best foot and spinal health guidance for users.

[0114] It should be understood that although the steps in the flowcharts involved in the above-described embodiments are displayed sequentially according to the indications of the arrows, these steps are not necessarily executed sequentially in the order indicated by the arrows. Unless there is a clear indication in this article, there is no strict order limit for the execution of these steps, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-described embodiments may include multiple steps or multiple stages. These steps or stages are not necessarily executed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be executed alternately or in turn with at least a part of other steps or steps or stages in other steps.

[0115] The above-described embodiments only represent several implementation manners of the embodiments of the present application. The description is relatively specific and detailed, but it should not be construed as a limitation on the scope of the patent of the embodiments of the application. It should be noted that for those of ordinary skill in the art, without departing from the concept of the embodiments of the present application, several modifications and improvements can still be made, and these all belong to the protection scope of the embodiments of the present application.

Claims

1. The full-cycle management method for adolescent foot spine health supported by the cloud platform is characterized by: The method comprises: Acquire the user's plantar pressure distribution data, spinal posture data and user behavior data in real time, wherein the user behavior data includes the user's daily step count, exercise duration and training compliance score; Performing spatiotemporal feature extraction processing on the plantar pressure distribution data to obtain a foot pressure feature vector; performing stability index calculation processing on the spinal posture data to obtain a spinal stability index; fusing the foot pressure feature vector, the spinal stability index and the user behavior data to obtain a joint feature vector; Based on the joint feature vector, a dynamic training parameter optimization process is performed by a reinforcement learning algorithm to obtain a dynamic training parameter combination, wherein the dynamic training parameter combination includes training intensity, complexity and action variants; Based on the user's progress speed, the training cycle is adjusted through the proportional-integral-derivative control algorithm to obtain an adaptive training cycle; Based on the joint feature vector and the adaptive training cycle, a new action generation process is performed using a generative adversarial network to obtain a variant training action, and a similarity screening process is performed on the variant training action to obtain a low-similarity action set; Based on the low-similarity action set, the dynamic training parameter combination is associated with the 3D printed orthotic device parameters, and a personalized foot spine health management plan is generated through a preset cloud platform. The personalized foot spine health management plan includes a dynamically updated training plan, orthotic device adaptation parameters and a long-term tracking strategy.

2. The cloud platform-supported full-cycle management method for adolescent foot spine health according to claim 1 is characterized in that: The step of performing dynamic training parameter optimization processing based on the joint feature vector by using a reinforcement learning algorithm to obtain a dynamic training parameter combination includes: Based on the joint feature vector, a reinforcement learning state space and an action space definition process are performed to obtain a state space and an action space; Based on the state space and the action space, the reward function is calculated using the following formula to obtain a reward value: Reward value = 0.6 × ΔP target -0.3×ΔI s +0.1×C Among them, ΔP target Indicates the pressure improvement rate of the foot pressure feature vector in the target area, ΔI s represents the absolute value of the change in the spinal stability index, and C represents the training compliance score; Based on the reward value, a policy network parameter update process is performed using a deep deterministic policy gradient algorithm to obtain the dynamic training parameter combination.

3. The full-cycle management method for adolescent foot spine health supported by a cloud platform according to claim 1 is characterized in that: The method of adjusting the training cycle based on the user's progress speed by using a proportional-integral-differential control algorithm to obtain an adaptive training cycle includes: The user's progress rate is calculated based on the user behavior data using the following formula: Among them, V represents the user's learning progress speed, m represents the number of learning behavior records, and L i represents the i-th learning assessment score, α represents the improvement rate coefficient, β i represents the weight factor of the i-th learning, R represents the user's comprehensive progress speed, p represents the number of evaluation indicators, and E k represents the kth evaluation index score, W k represents the weight of the kth evaluation indicator; Based on the progress speed, the training cycle is adjusted by a proportional-integral-derivative control algorithm to obtain the adaptive training cycle.

4. The full-cycle management method for adolescent foot spine health supported by a cloud platform according to claim 1 is characterized in that: The method of generating a new action based on the joint feature vector and the adaptive training cycle using a generative adversarial network to obtain a variant training action, and performing similarity screening on the variant training action to obtain a low-similarity action set includes: Based on the joint feature vector and the adaptive training cycle, a new action generation process is performed by a generative adversarial network to obtain a variant training action; Based on the variant training action and the historical action parameter vector, a cosine similarity calculation process is performed to obtain a similarity value; Based on the similarity value, a threshold screening process is performed to obtain the low-similarity action collection.

5. The full-cycle management method for adolescent foot spine health supported by a cloud platform according to claim 1 is characterized in that: The step of performing spatiotemporal feature extraction processing on the plantar pressure distribution data to obtain a plantar pressure feature vector; performing stability index calculation processing on the spinal posture data to obtain a spinal stability index; and fusing the plantar pressure feature vector, the spinal stability index and the user behavior data to obtain a joint feature vector includes: The following formula is used to perform principal component analysis on the plantar pressure distribution data to obtain the plantar pressure feature vector: λν=Cν Y=W T (X-µ) Where C represents the covariance matrix, n represents the number of samples, and x i represents the data vector of the i-th plantar pressure sampling point, μ represents the mean vector of all sampling points, λ represents the eigenvalue, ν represents the eigenvector, Y represents the eigenvector after dimensionality reduction, W represents the eigenvector matrix, and X represents the original plantar pressure data matrix; Performing stability index calculation processing on the spinal posture data to obtain the spinal stability index; The foot pressure feature vector, the spinal stability index and the user behavior data are input into a trained autoencoder neural network for feature fusion processing to obtain the joint feature vector.

6. The cloud platform-supported full-cycle management method for adolescent foot spine health according to claim 1 is characterized in that: Based on the low-similarity action set, the dynamic training parameter combination is associated with the 3D printed orthotic device parameters, and a personalized foot spine health management plan is generated through a preset cloud platform, including: Based on the low-similarity action set and the dynamic training parameters, a principal component analysis algorithm is used to extract action spatiotemporal features to generate an action feature vector; Performing Euclidean distance matching processing on the motion feature vector and a preset 3D printing orthopedic device parameter database to obtain a target device parameter combination; Based on the target instrument parameter combination, a model is modeled by a multivariate linear regression algorithm to obtain a personalized foot spine health management model; According to the personalized foot spine health management model, the correlation coefficients of the dynamic training parameters and the 3D printed orthotic device parameters are calculated to obtain a parameter correlation matrix; Based on the parameter association matrix, an optimal matching degree calculation process is performed by a gradient descent algorithm to obtain adjusted orthopedic appliance parameters; The adjusted orthotic device parameters are fused with the dynamic training parameters to generate the personalized foot spine health management program.

Images