Spine brace optimization-oriented gait feature intelligent analysis method

Abstract

The invention discloses a spine brace optimization-oriented gait feature intelligent analysis method, and relates to the technical field of medical image analysis and biomedical engineering.The method comprises the following steps: extracting gait category parameters from gait data after data processing, splicing the gait category parameters into gait feature vectors, and carrying out gait feature extraction on the gait feature vectors; analyzing the gait feature vector by using a principal component analysis algorithm to obtain gait feature parameters; combining the gait feature parameters with predefined brace feature parameters to construct a deep neural network model, and inputting the real-time gait feature parameters into the deep neural network model to obtain predicted brace feature parameters; and constructing a spine individual gait feature library based on the gait feature parameters and the brace feature parameters. According to the method, the complexity and diversity of gaits are comprehensively reflected by constructing the multi-layer gait parameter set and fusing the multi-modal data, and the accuracy and robustness of gait analysis are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the field of medical image analysis and biomedical engineering technology, and in particular to a gait feature intelligent analysis method for spinal brace optimization. Background Art

[0002] In the treatment process of scoliosis, gait analysis, as a key auxiliary means, can reveal the abnormal gait caused by scoliosis. Through in-depth analysis of gait, the limitations of imaging in static assessment can be effectively compensated, thereby providing more accurate guidance for the optimal design of spinal braces. In order to create a spinal brace that better meets individual needs and has a significant correction effect, researchers have introduced multimodal data fusion technology to enhance the quality of gait analysis. These multimodal data cover three-dimensional motion capture data, electromyography data, and plantar pressure distribution data. However, how to effectively integrate and analyze these heterogeneous information from different sensors or data sources has become a problem that needs to be solved urgently.

[0003] In recent years, researchers have conducted in-depth explorations on the effective fusion of multi-source heterogeneous gait data. Some studies have successfully reduced unnecessary feature inputs and improved the efficiency of gait recognition through multi-feature fusion technology. Other studies have proposed a hierarchical feature-level fusion method, which significantly improved the overall recognition rate of the system by optimizing the proportion of each feature in gait feature fusion. In addition, some studies have combined key features extracted from gait sequence images and applied them to human recognition systems, enhancing the practicality and accuracy of the system. At the same time, there are also research teams that have integrated visual and tactile features to construct a full-body gait model that includes upper limb swing. This model shows higher robustness and gait recognition accuracy in complex gait motion scenarios.

[0004] Although these methods have achieved remarkable results, there are still some shortcomings:

[0005] (1) Feature redundancy and extraction difficulties. Since the data involved in gait analysis is rich in dimensions and has high-frequency characteristics, there are a large number of redundant or irrelevant features. In order to improve the efficiency and accuracy of the model, it is necessary to screen out the core features that are sensitive to gait variation and have a key impact on the personalized design of the brace from the massive data. However, how to efficiently identify and extract these key features is a complex and challenging problem.

[0006] (2) Poor interpretability and comprehensibility. With the development of feature fusion technology, the generated fusion features are often high-dimensional and complex, which to a certain extent reduces the transparency and interpretability of the model. Specifically in the field of spinal brace optimization design, it is difficult for researchers to intuitively understand the specific contribution of each fusion feature to the orthopedic effect, and how they work together in the brace design optimization process. In addition, there is currently a lack of systematic methodological guidance when determining the weight distribution between different data sources or features, which makes it difficult to formulate effective and easy-to-understand fusion strategies in practical applications.

[0007] Currently, no effective solution has been proposed for the problems in the related technologies. Summary of the invention

[0008] In view of the problems in the related art, the present invention proposes an intelligent gait feature analysis method for spinal brace optimization to overcome the above-mentioned technical problems existing in the existing related art.

[0009] To this end, the specific technical solution adopted by the present invention is as follows:

[0010] An intelligent gait characteristic analysis method for spinal brace optimization includes the following steps:

[0011] S1. Collecting the patient's gait data, and processing the gait data, extracting gait category parameters from the processed gait data, splicing the gait category parameters into a gait feature vector, and analyzing the gait feature vector using a principal component analysis algorithm to obtain gait feature parameters;

[0012] S2. Combining gait characteristic parameters with predefined brace characteristic parameters, constructing a deep neural network model, inputting the real-time gait characteristic parameters into the deep neural network model, obtaining predicted brace characteristic parameters, and optimizing the deep neural network model based on the brace characteristic parameters;

[0013] S3. Based on the gait characteristic parameters and the brace characteristic parameters, a spine individual gait characteristic library is constructed, new gait data is collected, and the spine individual gait characteristic library is updated using the new gait characteristic data.

[0014] Further, collecting the patient's gait data, and processing the gait data, extracting gait category parameters from the processed gait data, splicing the gait category parameters into a gait feature vector, and analyzing the gait feature vector using a principal component analysis algorithm to obtain the gait feature parameters include the following steps:

[0015] S11, using several groups of sensors to collect gait data of the patient, and synchronously processing the gait data collected by each sensor through a timestamp;

[0016] S12, preprocessing and data optimization of the synchronously processed gait data;

[0017] S13, extracting gait category parameters from the optimized gait data, and constructing a gait parameter set based on the gait category parameters;

[0018] S14, concatenating the gait category parameters in the gait parameter set to obtain a gait feature vector;

[0019] S15. Combining the dynamic principal component analysis algorithm with the multimodal principal component analysis algorithm, the gait feature vector is analyzed to obtain gait feature parameters.

[0020] Furthermore, preprocessing and data optimization of the synchronized gait data include the following steps:

[0021] S121, performing data cleaning on the gait data after synchronization processing, wherein the data cleaning includes processing missing values, removing outliers and data verification;

[0022] S122, using a high-pass filter to perform denoising on the gait data after data cleaning, and performing standardization on the denoising gait data;

[0023] S123. Based on the standardized gait data, the time series data is divided into time windows of preset length through a recursive neural network model, the continuous gait data is processed using a sliding window technique, and the recursive neural network model parameters are updated through a gradient descent algorithm.

[0024] Furthermore, based on the standardized gait data, the time series data is divided into time windows of preset lengths through a recursive neural network model, the continuous gait data is processed using a sliding window technique, and the recursive neural network model parameters are updated through a gradient descent algorithm, including the following steps:

[0025] S1231, initializing the weight and learning rate of the recursive neural network model based on the standardized gait data;

[0026] S1232, using a sliding window technique, dividing the time series data into time windows of a preset length, reading gait data points from the standardized gait data stream, adding the gait data points to the time window, and extracting all gait data points within the time window;

[0027] S1233, based on each gait data point in the time window, executing the forward propagation process of the recursive neural network model, and combining the back propagation algorithm to calculate the gradient of the loss function to the recursive neural network model parameters;

[0028] S1234, combining the calculated gradient with a preset learning rate, and updating the parameters of the recursive neural network model using a gradient descent algorithm;

[0029] S1235. Repeat steps S1232 to S1234 to read new gait data points from the gait data stream until the gait data stream processing is completed.

[0030] Furthermore, combining the dynamic principal component analysis algorithm with the multimodal principal component analysis algorithm, analyzing the gait feature vector, and obtaining the gait feature parameters includes the following steps:

[0031] S151, performing singular value decomposition on the gait feature vector to obtain an initial principal component matrix;

[0032] S152, in each time window, using a dynamic principal component analysis algorithm and combining it with a recursive algorithm to update the principal component matrix, projecting the gait data through the updated principal component matrix, and extracting dynamic principal components;

[0033] S153, based on the extracted dynamic principal components, using a multimodal principal component analysis algorithm, defining a joint optimization objective function, and optimizing the principal component matrix by an alternating minimization algorithm;

[0034] S154. Use the optimized principal component matrix to extract features from the gait data to obtain gait feature parameters.

[0035] Furthermore, the expression of optimizing the principal component matrix by the alternating minimization algorithm is:

[0036]

[0037] Where U represents the shared principal component matrix; M represents the number of modes; m represents the index value; X m represents the gait data matrix of the mth mode; V m represents the specific principal component matrix of the mth mode; T represents the transpose of the matrix; ||.|| F represents the Frobenius norm.

[0038] Furthermore, combining the gait characteristic parameters with the predefined brace characteristic parameters, constructing a deep neural network model, inputting the real-time gait characteristic parameters into the deep neural network model, obtaining the predicted brace characteristic parameters, and optimizing the deep neural network model based on the brace characteristic parameters includes the following steps:

[0039] S21, combining gait characteristic parameters with predefined brace characteristic parameters to establish a characteristic parameter set;

[0040] S22. Based on the feature parameter set, a deep neural network model is constructed using a deep neural network algorithm, and the deep neural network model is trained using the gait feature parameters;

[0041] S23, collecting real-time gait characteristic parameters, performing data processing on the real-time gait characteristic parameters, extracting real-time gait category parameters, and obtaining real-time gait characteristic parameters based on the real-time gait category parameters;

[0042] S24, inputting the real-time gait characteristic parameters into the trained deep neural network model to obtain predicted brace characteristic parameters;

[0043] S25. Based on the predicted brace characteristic parameters, use the controller to adjust the predefined brace characteristic parameters, collect feedback information, and optimize the deep neural network model based on the feedback information.

[0044] Furthermore, based on the feature parameter set, a deep neural network model is constructed using a deep neural network algorithm, and the deep neural network model is trained using the gait feature parameters, including the following steps:

[0045] S221. Based on the feature parameter set, using a deep neural network algorithm, construct a deep neural network model including an input layer, a hidden layer, and an output layer, and establish a mapping relationship between gait feature parameters and brace feature parameters;

[0046] S222, using the mean square error as the loss function of the deep neural network model, and randomly initializing the parameters of the deep neural network model;

[0047] S223, inputting the gait characteristic parameters into the deep neural network model, obtaining the predicted brace characteristic parameters through the forward propagation algorithm, and calculating the loss value using the loss function according to the predicted brace characteristic parameters and the actual brace characteristic parameters;

[0048] S224, based on the loss value, using the back propagation algorithm to calculate the gradient of each parameter, and update the parameters based on the gradient;

[0049] S225. Repeat steps S223 to S224 until a preset number of iterations is reached.

[0050] Furthermore, the expression of the controller is:

[0051]

[0052] Where u(t) represents the control output at time t; K p represents the proportional gain; e(t) represents the error between the predicted brace characteristic parameters and the actual brace characteristic parameters; K irepresents the integral gain; t represents the proportion; τ represents the integral variable in the integral term; K d Represents the differential gain.

[0053] Furthermore, based on the gait characteristic parameters and the brace characteristic parameters, constructing a spine individual gait characteristic library, collecting new gait data, and updating the spine individual gait characteristic library using the new gait characteristic data includes the following steps:

[0054] S31, establishing a gait feature table based on the gait feature parameters;

[0055] S32. Establishing a brace feature table based on the brace feature parameters;

[0056] S33, constructing a spine individual gait feature library according to the gait feature table and the brace feature table, and in combination with the patient's feedback information;

[0057] S34, using the stream processing platform to collect new gait data in real time, and setting the data collection frequency and data format based on the new gait data;

[0058] S35, using a stream processing framework to perform data cleaning, denoising and standardization processing on the new gait data, and extracting new gait feature parameters based on the processed new gait data;

[0059] S36, storing the new gait feature parameters into the spine individual gait feature library, and setting data sharding and replication strategies.

[0060] The beneficial effects of the present invention are:

[0061] 1. The present invention comprehensively reflects the complexity and diversity of gait by constructing a multi-level gait parameter set and multimodal data fusion, significantly improving the accuracy and robustness of gait analysis. Dynamic principal component analysis and multimodal principal component analysis extract the dynamic principal components of gait data, reduce data dimensions, retain key features, and improve the efficiency and accuracy of the model.

[0062] 2. The present invention introduces a deep neural network to establish a mapping relationship between gait characteristics and brace parameters, thereby realizing end-to-end intelligent conversion from user gait data to brace parameters and providing a personalized brace adjustment solution. Through anomaly detection and fault-tolerant mechanisms, abnormal patterns in gait data are identified to ensure the safety and reliability of the system. At the same time, by collecting professional opinions from clinicians and feedback from patients, the model is continuously optimized to ensure the performance and reliability of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0064] Figure 1 is a flow chart of a gait characteristic intelligent analysis method for spinal brace optimization according to an embodiment of the present invention;

[0065] Figure 2 is a flow chart of an embodiment of a gait characteristic intelligent analysis method for spinal brace optimization according to an embodiment of the present invention;

[0066] Figure 3 is a schematic diagram of gait feature extraction based on dynamic principal component analysis of a gait feature intelligent analysis method for spinal brace optimization according to an embodiment of the present invention;

[0067] Figure 4 The invention relates to the construction of a gait parameter set of a gait characteristic intelligent analysis method for spinal brace optimization according to an embodiment of the present invention;

[0068] Figure 5 It is a brace feature parameter set of a gait feature intelligent analysis method for spinal brace optimization according to an embodiment of the present invention. DETAILED DESCRIPTION

[0069] To further illustrate each embodiment, the present invention provides drawings, which are part of the disclosure of the present invention and are mainly used to illustrate the embodiments. They can be used in conjunction with the relevant descriptions in the specification to explain the operating principles of the embodiments. With reference to these contents, ordinary technicians in this field should be able to understand other possible implementation methods and advantages of the present invention.

[0070] According to an embodiment of the present invention, a gait characteristic intelligent analysis method for spinal brace optimization is provided.

[0071] The present invention is further described with reference to the accompanying drawings and specific embodiments. Figure 1-Figure 5 As shown, according to the gait characteristic intelligent analysis method for spinal brace optimization according to an embodiment of the present invention, the method comprises the following steps:

[0072] S1. Collect the patient's gait data, process the gait data, extract gait category parameters from the processed gait data, splice the gait category parameters into a gait feature vector, analyze the gait feature vector using a principal component analysis algorithm, and obtain gait feature parameters.

[0073] Preferably, collecting the patient's gait data, and processing the gait data, extracting gait category parameters from the processed gait data, splicing the gait category parameters into a gait feature vector, and analyzing the gait feature vector using a principal component analysis algorithm to obtain the gait feature parameters includes the following steps:

[0074] S11. Collect the patient's gait data using several groups of sensors, and synchronously process the gait data collected by each sensor through a timestamp.

[0075] It should be noted that various sensors (such as inertial measurement units (IMUs), pressure sensors, heart rate sensors, etc.) and wearable devices (such as smart watches, health bracelets, etc.) are used to collect gait data, which includes but is not limited to three-dimensional motion capture data, ground reaction force data, electromyography data, heart rate, body temperature and blood oxygen saturation data, step length and step frequency data, foot pressure distribution, joint angle changes, etc. Data from different sources are synchronized through timestamps to ensure the consistency and accuracy of multi-source data, and appropriate transmission technologies (such as Bluetooth 5.0, Wi-Fi6, etc.) are selected to achieve efficient and stable data transmission.

[0076] S12. Preprocessing and data optimization are performed on the synchronously processed gait data.

[0077] Preferably, preprocessing and data optimization of the synchronously processed gait data includes the following steps:

[0078] S121. Perform data cleaning on the synchronously processed gait data, where the data cleaning includes processing missing values, removing outliers, and verifying the data.

[0079] It should be noted that the data cleaning of the synchronized gait data includes:

[0080] Handling missing values: For missing values, choose to delete the records containing missing values, or use interpolation methods (such as linear interpolation, mean interpolation, etc.) to fill in the missing values.

[0081] Removing outliers: Calculate the mean and standard deviation of each gait data, and consider values ​​outside a certain range (such as mean ± 3 times standard deviation) as outliers. You can choose to delete outliers or replace them with neighboring values ​​or the mean.

[0082] Data verification (checking data consistency and integrity): Ensure that all data formats are consistent, such as timestamps, units, etc., and check whether the data has logical errors, such as whether the timestamps are arranged in order, whether the values ​​are reasonable, etc.

[0083] S122, using a high-pass filter to perform denoising on the gait data after data cleaning, and performing standardization on the denoising gait data.

[0084] It should be noted that a high-pass Butterworth filter is used to remove noise. The Butterworth filter is a commonly used filter with simple design and stable performance. Its main characteristics are flat frequency response in the passband, smooth transition band, and no ripple. The high-pass Butterworth filter is used to remove low-frequency noise and retain high-frequency signals.

[0085] The denoised gait data is standardized to make its mean 0 and variance 1. The formula is:

[0086]

[0087] In the formula, represents the original gait data matrix of the i-th mode; μ (i) represents the mean vector; σ (i) represents the standard deviation vector.

[0088] S123. Based on the standardized gait data, the time series data is divided into time windows of preset length through a recursive neural network model, the continuous gait data is processed using a sliding window technique, and the recursive neural network model parameters are updated through a gradient descent algorithm.

[0089] It should be noted that the formula for dividing time series data into time windows of preset length is:

[0090]

[0091] In the formula, represents the gait data of the i-th mode at time point t; t represents the current time point; w represents the length of the time window, and the optimal value usually needs to be determined through experiments.

[0092] Preferably, based on the standardized gait data, the time series data is divided into time windows of preset lengths through a recursive neural network model, the continuous gait data is processed using a sliding window technique, and the recursive neural network model parameters are updated through a gradient descent algorithm, including the following steps:

[0093] S1231. Initialize the weights and learning rate of the recursive neural network model based on the standardized gait data.

[0094] It should be noted that the initial weights are set using Xavier / Glorot initialization for each layer in the recursive neural network model. Xavier / Glorot initialization is a method for initializing neural network weights. This initialization method aims to keep the variance of the activation value and gradient of each layer roughly the same during forward and backward propagation, thereby helping to solve the problem of gradient vanishing or gradient exploding in deep networks.

[0095] Setting the learning rate is to choose a suitable initial learning rate and use the Adam optimizer to automatically adjust the learning rate. The Adam optimizer is a commonly used stochastic gradient descent optimization algorithm that combines the advantages of AdaGrad and RMSProp; Adam dynamically adjusts the learning rate of each parameter by calculating the first-order moment estimate and second-order moment estimate of the gradient. This adaptive learning rate method makes Adam very effective when processing large-scale data sets and parameter spaces.

[0096] S1232. Using sliding window technology, the time series data is divided into time windows of preset length, and gait data points are read from the standardized gait data stream, the gait data points are added to the time window, and all gait data points in the time window are extracted.

[0097] It should be noted that the definition of the time window size is to select an appropriate time window size according to the characteristics of the gait cycle; for example, it can be sensor data within a few seconds. Reading and filling the time window is to read data points from the standardized gait data stream in sequence and add these data points to the current time window. When the window is filled, start processing the data in this window.

[0098] S1233. Based on each gait data point in the time window, execute the forward propagation process of the recursive neural network model, and combine the back propagation algorithm to calculate the gradient of the loss function to the parameters of the recursive neural network model.

[0099] It should be noted that the forward propagation is to perform the forward propagation process for the gait data points in each time window, that is, to make predictions through the RNN model. The loss function is to select an appropriate loss function to measure the difference between the model prediction value and the actual label. For classification tasks, cross entropy is usually used; for regression tasks, mean square error can be used.

[0100] S1234. Combining the calculated gradient with the preset learning rate, the parameters of the recursive neural network model are updated using a gradient descent algorithm.

[0101] It should be noted that back propagation is to calculate the gradient of the loss function to the model parameters through the back propagation algorithm. Parameter update is to update the model parameters using the gradient descent algorithm and a predetermined learning rate to reduce the value of the loss function.

[0102] S1235. Repeat steps S1232 to S1234 to read new gait data points from the gait data stream until the gait data stream processing is completed.

[0103] It should be noted that data optimization includes two important steps: model parameter updating and continuous data stream processing. The gradient descent algorithm is used to update the model parameters in real time to adapt to data changes, so that the model maintains accuracy in a changing environment. The specific steps are as follows:

[0104] Step 1: Initialize model parameters:

[0105] The initial parameter vector θ0 of the model; the learning rate η is used to control the step size of each parameter update.

[0106] Step 2: Data flow definition:

[0107] Gait data points (x t ,y t ), where x t represents the feature vector at time t, y t Represents the target value at time t.

[0108] Eigenvector x t Including various gait data collected, such as step length, step frequency, joint angle changes, etc. Target value y t is the gait classification label.

[0109] Step 3: Calculate the gradient:

[0110] The loss function L(θ) is related to the parameter θ t The gradient of

[0111] The loss function L(θ) is used to measure the difference between the model prediction value and the actual value. For example, the mean square error (MSE) is used to calculate L(θ), and the formula is:

[0112] L(θ)=(h θ (x t )-y t ) 2 / 2;

[0113] Where L(θ) represents the loss function; h θ (x t ) represents the predicted value of the model; y t Indicates actual value.

[0114] For L(θ), the gradient is: Among them, θ t represents the model parameter vector at time point t;

[0115] Step 4: Update model parameters:

[0116] Update the model parameters θ using gradient descent t , the formula is:

[0117]

[0118] In the formula, θ t represents the model parameter vector at time point t; θ t+1 Represents the model parameter vector at time point t+1, that is, the parameters after a gradient descent update; Represents the gradient of the loss function at time point t.

[0119] Step 5: Repeat the process:

[0120] Repeat steps 2 to 4 until all data points have been processed.

[0121] Use sliding window technology to process continuous gait data stream to ensure real-time performance. The sliding window can be fixed in size or dynamically adjusted. The specific steps are as follows:

[0122] Step 1: Initialization:

[0123] (1) Define window size: Select an appropriate window size w, which is the number of gait data points.

[0124] (2) Initialize the window: Create an empty window list window to store the most recent w gait data points.

[0125] (3) Initialize model parameters: Initialize model parameters θ, which can be set to zero vector or random value; θ is a vector containing all parameters, including bias term θ0 and other feature parameters θ1, θ2, …, θ n .

[0126] (4) Set the learning rate: Select an appropriate learning rate α.

[0127] Step 2: Gait data stream processing:

[0128] Read gait data points one by one from the gait data stream (x t ,y t ).

[0129] The current gait data point (x t ,y t)Add window window, if the window size exceeds the window size w, remove the earliest data point.

[0130] Step 3: Batch gradient descent:

[0131] When the window contains w gait data points, the following steps are performed:

[0132] (1) Extract all gait data points in the window (x (i) ,y (i) ), where i = 1, 2, ..., w; x (i) represents the input feature vector of the i-th training sample; y (i) Represents the actual output value of the i-th training sample.

[0133] (2) Calculate the model's predicted value (x (i) ,y (i) ), calculate the predicted value h of the model θ (x (i) ), the calculation formula is as follows:

[0134] h θ (x (i) )=θ0+θ1x1 (i) +θ2x2 (i) +...+θ n x n (i) ;

[0135] In the formula, h θ (x (i) ) represents the predicted value of the model; n represents the number of parameters; x1 (i) , x2 (i) , …, x3 (i) represents the eigenvalues ​​of the i-th training sample; θ0, θ1, θ2, ..., θ n Represents the parameters of the model.

[0136] (3) Calculate each parameter θ j The gradient of is:

[0137]

[0138] In the formula, gradient j represents the gradient value; w represents the window size; h θ (x (i) ) represents the model's predicted value for the i-th training sample; y (i) is the actual output value of the i-th training sample; x j (i) is the j-th eigenvalue of the ith training sample.

[0139] (4) Update each parameter θ' using the learning rate α j ,Right now:

[0140] θ' j =θ j -α.gradient j ;

[0141] In the formula, gradient j represents the gradient value; α represents the learning rate; θ j Represents the parameter; θ' j Indicates the updated parameters.

[0142] Step 4: Repeat the process:

[0143] Return to step 2 and continue processing new data points until the data stream ends.

[0144] By using the gradient descent algorithm for real-time processing, we ensure that data can be pre-processed immediately during gait data collection, reducing latency and thus improving the performance of the entire system and user experience.

[0145] S13. Extracting gait category parameters from the optimized gait data, and constructing a gait parameter set based on the gait category parameters.

[0146] It should be noted that the gait category parameters are extracted from the optimized gait data, and the gait parameter set is constructed based on the gait category parameters, including:

[0147] Step 1: Construction of gait parameter set:

[0148] The multi-level construction of gait parameter sets can better reflect the complexity and diversity of gait, thereby improving the accuracy and robustness of gait analysis. The steps for constructing gait parameter sets are as follows:

[0149] (1) Construct a large set of gait feature parameter sets;

[0150] The major categories of gait characteristic parameters include:

[0151] C1: gait cycle, a complete cycle of gait, C1 = T 支撑相 +T 摆动相 ;

[0152] C2: step length, i.e. the distance of one step C2 = C2.1 + C2.2;

[0153] C3: Cadence, the distance of one step, equal to, total step length / time;

[0154] C4: step width, the distance between the center points of the two feet, C4 = (C4.1 + C4.2) / 2;

[0155] C5: pace, walking speed per unit time, equal to total distance / time;

[0156] C6: Joint angle changes, angle changes of key joints (such as ankle, knee, hip) during the gait cycle;

[0157] C7: Foot pressure distribution.

[0158] (2) Refine the feature parameters of large categories into feature parameters of small categories;

[0159] Refining the gait cycle:

[0160] C1.1: Stance phase ratio, the proportion of the stance phase in the entire gait cycle, C1.1 = T 支撑相 / C1;

[0161] C1.2: Swing phase ratio, the proportion of the swing phase in the entire gait cycle, C1.2 = T 摆动相 / C1.

[0162] Refinement step size:

[0163] C2.1: left step length, the step length of the left foot;

[0164] C2.2: Right step length, the step length of the right foot.

[0165] Refine your cadence:

[0166] C3.1: left cadence, the cadence of the left foot, equal to, number of left steps / time;

[0167] C3.2: Right cadence, the cadence of the right foot, equals right steps / time.

[0168] Refinement step width:

[0169] C4.1: Forefoot distance, the lateral distance between the front and rear feet;

[0170] C4.2: Backfoot distance, the lateral distance between the backfoot and the front foot.

[0171] Refine your pace:

[0172] C5.1: instantaneous pace, the instantaneous speed of each step;

[0173] C5.2: Average gait speed, the average speed of the entire gait cycle.

[0174] Refine the joint angle changes:

[0175] C6.1: Ankle angle, the angle change of the ankle joint during the gait cycle;

[0176] C6.2: knee angle, the angle change of the knee joint during the gait cycle;

[0177] C6.3: Hip angle, the angle of the hip joint during the gait cycle.

[0178] Refined foot pressure distribution:

[0179] C7.1: Forefoot pressure, pressure distribution of the forefoot at different gait phases;

[0180] C7.2: Heel pressure, distribution of heel pressure during different gait phases.

[0181] (3) Construction of gait parameter set;

[0182] Inheritance category features:

[0183] The major features C1, C2, C3, C4, C5, C6, and C7 are taken as basic features.

[0184] Combined with sub-category features:

[0185] The refined small class feature parameters C1.1, C1.2, C2.1, C2.2, C3.1, C3.2, C4.1, C4.2, C5.1, C5.2, C6.1, C6.2, C6.3, C7.1, C7.2 are combined into the overall features.

[0186] Construct gait parameters:

[0187] F1 = [C1, C1.1, C1.2];

[0188] F2 = [C2, C2.1, C2.2];

[0189] F3 = [C3, C3.1, C3.2];

[0190] F4 = [C4, C4.1, C4.2];

[0191] F5 = [C5, C5.1, C5.2];

[0192] F6=[C6, C6.1, C6.2, C6.3];

[0193] F7 = [C7, C7.1, C7.2].

[0194] Store parameter values ​​in a structured DataFrame, which is a common data structure. DataFrame has good readability and interpretability, and you can directly view and understand the value of each feature and its corresponding relationship.

[0195] S14. Concatenate the gait category parameters in the gait parameter set to obtain a gait feature vector.

[0196] It should be noted that the formula for concatenating the gait category parameters in the gait parameter set is:

[0197] X t =[X t (1) ,X t (2) ,…,X t (m) ];

[0198] Where, X t represents the gait category parameter; m represents the number of gait features; X t (1) , X t (2) , ..., X t (m) Corresponding to F1, F2, ..., F7 respectively.

[0199] S15. Combining the dynamic principal component analysis algorithm with the multimodal principal component analysis algorithm, the gait feature vector is analyzed to obtain gait feature parameters.

[0200] Preferably, combining the dynamic principal component analysis algorithm with the multimodal principal component analysis algorithm to analyze the gait feature vector and obtain the gait feature parameters includes the following steps:

[0201] S151. Perform singular value decomposition on the gait feature vector to obtain an initial principal component matrix.

[0202] It should be noted that the formula for obtaining the initial principal component matrix by performing singular value decomposition on the gait feature vector is:

[0203]

[0204] Wherein, U0 represents the left singular vector matrix; S0 represents the singular value matrix; V0 represents the right singular vector matrix; svd represents a matrix decomposition method in linear algebra, which can decompose the matrix X into the product of three matrices U0, S0 and V0.

[0205] S152. In each time window, a dynamic principal component analysis algorithm is used in combination with a recursive algorithm to update a principal component matrix, and the gait data is projected through the updated principal component matrix to extract dynamic principal components.

[0206] It should be noted that in each time window, the dynamic principal component analysis algorithm is used in combination with the recursive algorithm to update the principal component matrix, namely:

[0207] U t ,S t ,V t=svd(X t );

[0208]

[0209] Where U t represents the left singular vector matrix; S t represents the singular value matrix; V t represents the right singular vector matrix; svd represents a matrix decomposition method in linear algebra; X t represents the input data matrix at time point t; U t+1 represents the updated left singular vector matrix; Represents the new principal component of the current time window; α represents the learning rate, which controls the speed of updating and needs to be adjusted according to the rate of change of the data and the convergence of the model.

[0210] Use the updated principal components to project the gait data, namely:

[0211] Z t =X t *U t ;

[0212] In the formula, Z t represents the extracted dynamic principal component; X t represents the input data matrix at time point t; U t Represents the principal component vector matrix at time point t.

[0213] S153. Based on the extracted dynamic principal components, a multimodal principal component analysis algorithm is used to define a joint optimization objective function, and the principal component matrix is ​​optimized through an alternating minimization algorithm.

[0214] Preferably, the expression for optimizing the principal component matrix by the alternating minimization algorithm is:

[0215]

[0216] Where U represents the shared principal component matrix; M represents the number of modes; m represents the index value; X m represents the gait data matrix of the mth mode; V m represents the specific principal component matrix of the mth mode; T represents the transpose of the matrix; ||.|| F represents the Frobenius norm.

[0217] It should be noted that a joint optimization objective function is defined to minimize the reconstruction error of each modal data in the principal component space and ensure the synergy of different modal data, as follows:

[0218]

[0219] Where U represents the shared principal component matrix, which has a shape of d×k, where d represents the feature dimension and k represents the number of principal components; V m Represents the specific principal component matrix of the mth mode, with a shape of k×n m , where n m represents the number of samples of the mth mode; X m The data matrix representing the mth mode has a shape of d×n m ;||.|| F represents the Frobenius norm, which is the square root of the sum of the squares of the matrix elements.

[0220] The objective function is optimized by the alternating minimization method and U is updated, namely:

[0221]

[0222] Where U represents the shared principal component matrix; M represents the number of modes; m represents the index value; X m represents the gait data matrix of the mth mode; V m represents the specific principal component matrix of the mth mode; T represents the transpose of the matrix; ||.|| F represents the Frobenius norm.

[0223] Update V m ,Right now:

[0224]

[0225] Where V m represents the updated principal component matrix; X m represents the gait data matrix of the mth mode; V m represents the specific principal component matrix of the mth mode; T represents the transpose of the matrix; ||.|| F represents the Frobenius norm.

[0226] Among them, V m T Indicates V m The transposed matrix of The superscript 2 indicates square, and argmin indicates the parameter value that minimizes the objective function.

[0227] S154. Use the optimized principal component matrix to extract features from the gait data to obtain gait feature parameters.

[0228] It should be noted that the gait data is extracted using the optimized principal component matrix, and the gait characteristic parameters obtained include:

[0229] Z m =X m .U;

[0230] Among them, Z m Represents the gait feature extraction result of the mth mode, and the shape represents k×n m , where k represents the number of principal components, X m Represents the original data matrix of the mth mode, whose shape is d×n m , where d represents the feature dimension; U represents the shared principal component matrix, and the shape represents d×k, where d represents the feature dimension and k represents the number of principal components.

[0231] In addition, the dynamic principal component analysis algorithm and the multimodal principal component analysis algorithm are combined to effectively extract the characteristics of gait data, which not only considers the temporal correlation but also makes full use of the complementary information of different modal data, thereby improving the accuracy and robustness of gait analysis.

[0232] S2. Combining the gait characteristic parameters with the predefined brace characteristic parameters, constructing a deep neural network model, inputting the real-time gait characteristic parameters into the deep neural network model, obtaining the predicted brace characteristic parameters, and optimizing the deep neural network model based on the brace characteristic parameters.

[0233] It should be noted that the predefined brace characteristic parameters include:

[0234] (1)Material properties;

[0235] Hardness: The hardness value of the brace material (Shore A or Shore D), marked as H1.

[0236] (2) Shape design;

[0237] Concave or convex design: Boolean value, whether it is designed to be slightly convex in the front and slightly concave in the back to adapt to the natural curve of the spine, mark: SD1.

[0238] Side wing design: Boolean value, whether to design wing-like structures on both sides to increase lateral support and prevent lateral bending, mark: SD2.

[0239] Lumbar support: Boolean value, whether to design enhanced support in the lumbar area to reduce the burden on the lumbar spine, mark: SD3.

[0240] Thoracic support: Boolean value, whether to design appropriate support in the thoracic spine area to maintain the normal curvature of the thoracic spine, mark: SD4.

[0241] Cervical support: Boolean value, whether to design gentle support in the cervical area to relieve neck pressure, mark: SD5.

[0242] (3) Dimensional parameters;

[0243] a. Length;

[0244] Length of cervical vertebrae, labeled: L1;

[0245] Thoracic segment length, labeled: L2;

[0246] Lumbar segment length, labeled: L3;

[0247] The sacral segment is long and is labeled: L4.

[0248] b. Width;

[0249] Shoulder width, marked: W1;

[0250] Waist width, marked: W2.

[0251] c. Thickness;

[0252] Thickness of the brace, marked: T1.

[0253] d. Regulatory mechanism;

[0254] Elasticity adjustment: Boolean value, whether it is equipped with an adjustable elastic band or Velcro so that the user can adjust the tightness as needed, mark: RM1.

[0255] Height adjustment: Boolean value, whether to provide different supports at different heights, flag: RM2.

[0256] e. Breathability;

[0257] Breathability: Boolean value, whether the material is breathable, flag: B1.

[0258] Number of ventilation holes: Design ventilation holes at key locations of the brace to increase air circulation, marked: B2.

[0259] f. Functional parameters;

[0260] Pressure distribution: Boolean value, whether to design a reasonable pressure distribution to avoid discomfort caused by excessive local pressure, mark: FP1.

[0261] Temperature Regulation: Boolean value, whether to use temperature control materials, such as phase change materials (PCM), to regulate the temperature inside the brace, flag: FP2.

[0262] Sensor Integration: Boolean value, whether to integrate sensors such as pressure sensors, temperature sensors, etc. to monitor the user's gait and posture, tag: FP3.

[0263] The above brace feature parameters are expressed as a feature set Y, expressed as:

[0264] Y={H1, SD1, SD2, SD3, SD4, SD5, L1, L2, L3, L4, W1, W2, T1, RM1, RM2, B1, B2, FP1, FP2, FP3}.

[0265] Preferably, combining gait characteristic parameters with predefined brace characteristic parameters to construct a deep neural network model, inputting the real-time gait characteristic parameters into the deep neural network model to obtain predicted brace characteristic parameters, and optimizing the deep neural network model based on the brace characteristic parameters includes the following steps:

[0266] S21. Combining gait characteristic parameters with predefined brace characteristic parameters to establish a characteristic parameter set.

[0267] It should be noted that the brace feature set Y = {y1, y2, ..., y k};y k represents the predefined brace characteristic parameters; k represents the number of brace characteristic parameters.

[0268] Set the gait feature set X = {x1, x2, ..., x n}; x n represents the gait characteristic parameters; n is the number of gait characteristic parameters.

[0269] S22. Based on the feature parameter set, a deep neural network model is constructed using a deep neural network algorithm, and the deep neural network model is trained using the gait feature parameters.

[0270] Preferably, based on the feature parameter set, a deep neural network model is constructed using a deep neural network algorithm, and the deep neural network model is trained using gait feature parameters, including the following steps:

[0271] S221. Based on the feature parameter set and using a deep neural network algorithm, a deep neural network model including an input layer, a hidden layer, and an output layer is constructed to establish a mapping relationship between gait feature parameters and brace feature parameters.

[0272] S222. Use the mean square error as the loss function of the deep neural network model, and randomly initialize the parameters of the deep neural network model.

[0273] S223. Input the gait characteristic parameters into the deep neural network model, obtain the predicted brace characteristic parameters through the forward propagation algorithm, and calculate the loss value using the loss function based on the predicted brace characteristic parameters and the actual brace characteristic parameters.

[0274] S224. Based on the loss value, the gradient of each parameter is calculated using the back propagation algorithm, and the parameters are updated based on the gradient.

[0275] S225. Repeat steps S223 to S224 until a preset number of iterations is reached.

[0276] It should be noted that the deep neural network algorithm is used to introduce the deep neural network (DNN) as the basic model to establish the mapping relationship f: X→Y, which is specifically expressed as: Y=f(X)+∈, where ∈ represents the error term; Y represents the brace characteristic parameter; X represents the gait characteristic parameter; and f represents the mapping relationship.

[0277] The structures of the deep neural network (DNN) are as follows:

[0278] Input layer: receives the feature vector in the gait feature set X.

[0279] Hidden layer: Contains multiple layers of fully connected layers, each of which can introduce nonlinearity through activation functions (such as ReLU).

[0280] Output layer: Output the design parameters in the brace feature set Y.

[0281] The training goal of the deep neural network model is to minimize the loss function L, which uses the mean square error (MSE) and is formulated as follows:

[0282]

[0283] Where, L represents the loss function; N represents the number of training samples; i represents the index value; Y represents the brace feature parameter; X represents the gait feature parameter; f represents the mapping relationship; θ represents the parameter.

[0284] The deep neural network model training process is as follows:

[0285] (1) Initialize model parameters: Randomly initialize the model parameters θ.

[0286] (2) Forward propagation: Input data X is fed into the model to obtain the predicted brace features.

[0287] (3) Calculate the loss: Select mean square error (MSE) to measure the gap between the predicted value and the actual value.

[0288] (4) Back propagation: Update the model parameters θ through the back propagation algorithm to reduce the value of the loss function.

[0289] (5) Repeat training: Iterate the above process multiple times until the model converges or reaches the predetermined training rounds.

[0290] S23, collecting real-time gait characteristic parameters, performing data processing on the real-time gait characteristic parameters, extracting real-time gait category parameters, and obtaining real-time gait characteristic parameters based on the real-time gait category parameters.

[0291] It should be noted that the user's gait data is continuously collected, and the real-time gait features are extracted from the processed gait data to form the real-time gait feature parameters X'={x1', x2', ..., x n '}.

[0292] S24. Input the real-time gait characteristic parameters into the trained deep neural network model to obtain predicted brace characteristic parameters.

[0293] S25. Based on the predicted brace characteristic parameters, use the controller to adjust the predefined brace characteristic parameters, collect feedback information, and optimize the deep neural network model based on the feedback information.

[0294] Preferably, the expression of the controller is:

[0295]

[0296] Where u(t) represents the control output at time t; K p represents the proportional gain; e(t) represents the error between the predicted brace characteristic parameters and the actual brace characteristic parameters; K i represents the integral gain; t represents the proportion; τ represents the integral variable in the integral term; K d Represents the differential gain.

[0297] It should be noted that, according to the predicted brace characteristics Y', the brace parameters are automatically adjusted to provide personalized support and correction.

[0298] The controller is a PID controller, which is a commonly used feedback control system that adjusts the control output through three parts: proportional, integral and differential to achieve the desired control effect.

[0299] In addition, the professional opinions of clinicians and feedback from patients are collected to continuously optimize the deep neural network model. If the patient feels uncomfortable or has other abnormal conditions during use, the system should have fault tolerance and be able to respond in time to ensure safety.

[0300] Collecting professional opinions from clinicians and feedback from patients includes:

[0301] (1) Clinician feedback: Regularly collect the opinions and suggestions of clinicians, especially feedback on the comfort, support effect and patient experience of the brace; obtain the professional opinions of doctors through questionnaires, interviews or meetings.

[0302] (2) Patient feedback: Collect patients’ usage experience and feelings through user questionnaires, online feedback forms, or telephone follow-up; pay attention to any discomfort or abnormalities encountered by patients during use and record detailed information.

[0303] Continuous optimization of the model based on online learning includes:

[0304] (1) Data processing

[0305] (2) Online learning model update;

[0306] a. Initialize the model: Randomly initialize the model parameters θ0; select an appropriate learning rate η.

[0307] b. Gait data stream processing: read gait data points one by one from the gait data stream (x t ,y t , f t ), where x t represents the gait characteristic parameter, y t represents the characteristic parameter of the brace, f t Indicates feedback information.

[0308] c. Forward propagation: input gait feature parameters x t Enter the model to obtain the predicted brace characteristic parameters

[0309] d. Calculate the loss:

[0310] The loss function L is calculated using mean square error (MSE) and cross entropy loss (CE), namely:

[0311]

[0312] Where L represents the loss function; α represents the weight parameter, which is used to balance the importance of the two losses; Represents the predicted brace characteristic parameters; y i Indicates the characteristic parameters of the brace; i indicates the index value; Indicates the feedback information of the prediction; f t Indicates feedback information.

[0313] e. Back propagation:

[0314] Calculate the gradient through the back-propagation algorithm Update model parameters θ t to θ t+1 , calculated as follows:

[0315]

[0316] In the formula, θ t+1represents the updated parameters; θ t represents the parameter; η represents the learning rate; x t represents the gait characteristic parameter, y t represents the characteristic parameter of the brace, f t Indicates feedback information.

[0317] Anomaly detection and fault tolerance mechanisms include:

[0318] (1) Anomaly detection;

[0319] Identify abnormal patterns in gait data or abnormal changes in brace performance by setting thresholds or using anomaly detection algorithms. Use statistical methods (such as Z-score) or to identify abnormal patterns. The details are as follows:

[0320] Z = (x - μ) / σ;

[0321] In the formula, Z represents an outlier; x represents an observed value; μ represents the mean; and σ represents the standard deviation. If |Z|>3, it is considered an outlier.

[0322] Monitor the system's operating status to ensure that all sensors and actuators are functioning properly.

[0323] (2) Fault-tolerant response;

[0324] When an abnormal situation is detected, the system takes immediate action, such as suspending adjustments, sounding an alarm, or reverting to a default safe configuration. An emergency stop button is provided to allow manual intervention by the user or caregiver.

[0325] (3) Reset and recovery;

[0326] After the abnormal situation is resolved, the system can quickly return to normal working state and restart the monitoring and adjustment process.

[0327] S3. Based on the gait characteristic parameters and the brace characteristic parameters, a spine individual gait characteristic library is constructed, new gait data is collected, and the spine individual gait characteristic library is updated using the new gait characteristic data.

[0328] Preferably, constructing a spine individual gait feature library based on gait feature parameters and brace feature parameters, collecting new gait data, and updating the spine individual gait feature library using the new gait feature data includes the following steps:

[0329] S31, establishing a gait feature table based on the gait feature parameters;

[0330] S32. Establishing a brace feature table based on the brace feature parameters;

[0331] S33, constructing a spine individual gait feature library according to the gait feature table and the brace feature table, and in combination with the patient's feedback information;

[0332] S34, using the stream processing platform to collect new gait data in real time, and setting the data collection frequency and data format based on the new gait data;

[0333] S35, using a stream processing framework to perform data cleaning, denoising and standardization processing on the new gait data, and extracting new gait feature parameters based on the processed new gait data;

[0334] S36, storing the new gait feature parameters into the spine individual gait feature library, and setting data sharding and replication strategies.

[0335] It should be noted that, in a specific embodiment, Amazon DynamoDB, a database system that supports distributed storage and multimodal data, is selected, a database instance is created in the selected database system, and necessary security settings and backup policies are configured.

[0336] Step 1: Design the table structure;

[0337] (1) Individual information form:

[0338] Fields: Individual name, age, gender, height, weight, scoliosis type, treatment history, wearable device ID, environmental sensor data ID.

[0339] (2) Gait feature table:

[0340] Fields: timestamp of data collection, feature name (such as single-leg support period, stride size, cadence, etc.), feature value, sensor type (such as 3D motion capture, ground reaction force, electromyography, etc.), wearable device data, environmental sensor data.

[0341] (3) Subjective feedback form:

[0342] Fields: timestamp of feedback, pain score, comfort rating.

[0343] (4) Brace parameter table:

[0344] Fields: timestamp of parameter adjustment, parameter name (such as hardness, shape, size, etc.), parameter value, reason for adjustment (such as doctor's advice, etc.).

[0345] Step 2: Database maintenance and update;

[0346] (1) New gait data collection:

[0347] Use a stream processing platform to continuously collect and process new gait data from different sensors and devices, and configure data collection frequency and data format.

[0348] (2) Gait data preprocessing:

[0349] Use the stream processing framework to clean, denoise and standardize gait data in real time, and implement gait data verification and error handling logic.

[0350] (3) Feature extraction:

[0351] Extract key gait features in real time.

[0352] (4) Data storage:

[0353] Store the processed data in real time in the distributed database system Amazon DynamoDB, configure data sharding and replication strategies to ensure performance and reliability during high-concurrency access.

[0354] Data Sharding: Data sharding means dividing data into multiple fragments and distributing them on different nodes to improve read and write performance and scalability. For example, using hash sharding, that is, assigning data to different shards based on the hash value of a field in the data (such as individual_id).

[0355] Data replication: Data is replicated on multiple nodes to improve data availability and fault tolerance. For example, synchronous replication is used, that is, when writing data, the data is written to multiple nodes at the same time to ensure data consistency.

[0356] (5)Automation script:

[0357] Regularly check and optimize the performance of automation scripts to ensure efficient operation of the system.

[0358] In summary, with the help of the above technical solution of the present invention, by constructing a multi-level gait parameter set and multimodal data fusion, the complexity and diversity of gait are fully reflected, and the accuracy and robustness of gait analysis are significantly improved. Dynamic principal component analysis and multimodal principal component analysis extract the dynamic principal components of gait data, reduce data dimensions, retain key features, and improve the efficiency and accuracy of the model. By introducing a deep neural network to establish a mapping relationship between gait features and brace parameters, end-to-end intelligent conversion from user gait data to brace parameters is achieved, providing a personalized brace adjustment solution; through anomaly detection and fault-tolerant mechanisms, abnormal patterns in gait data are identified to ensure the safety and reliability of the system. At the same time, by collecting professional opinions from clinicians and feedback from patients, the model is continuously optimized to ensure the performance and reliability of the system.

[0359] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. An intelligent gait feature analysis method for spinal brace optimization, characterized in that: The method comprises the following steps: S1. Collecting the patient's gait data, and processing the gait data, extracting gait category parameters from the processed gait data, splicing the gait category parameters into a gait feature vector, and analyzing the gait feature vector using a principal component analysis algorithm to obtain gait feature parameters; S2. Combining gait characteristic parameters with predefined brace characteristic parameters, constructing a deep neural network model, inputting the real-time gait characteristic parameters into the deep neural network model, obtaining predicted brace characteristic parameters, and optimizing the deep neural network model based on the brace characteristic parameters; S3. Based on the gait characteristic parameters and the brace characteristic parameters, a spine individual gait characteristic library is constructed, new gait data is collected, and the spine individual gait characteristic library is updated using the new gait characteristic data.

2. The gait characteristic intelligent analysis method for spinal brace optimization according to claim 1, characterized in that: The step of collecting the patient's gait data, processing the gait data, extracting gait category parameters from the processed gait data, splicing the gait category parameters into a gait feature vector, and analyzing the gait feature vector using a principal component analysis algorithm to obtain the gait feature parameters comprises the following steps: S11, using several groups of sensors to collect gait data of the patient, and synchronously processing the gait data collected by each sensor through a timestamp; S12, preprocessing and data optimization of the synchronously processed gait data; S13, extracting gait category parameters from the optimized gait data, and constructing a gait parameter set based on the gait category parameters; S14, concatenating the gait category parameters in the gait parameter set to obtain a gait feature vector; S15. Combining the dynamic principal component analysis algorithm with the multimodal principal component analysis algorithm, the gait feature vector is analyzed to obtain gait feature parameters.

3. The gait characteristic intelligent analysis method for spinal brace optimization according to claim 2, characterized in that: The preprocessing and data optimization of the synchronously processed gait data comprises the following steps: S121, performing data cleaning on the gait data after synchronization processing, wherein the data cleaning includes processing missing values, removing outliers and data verification; S122, using a high-pass filter to perform denoising on the gait data after data cleaning, and performing standardization on the denoising gait data; S123. Based on the standardized gait data, the time series data is divided into time windows of preset lengths through a recursive neural network model, the continuous gait data is processed using a sliding window technique, and the recursive neural network model parameters are updated through a gradient descent algorithm.

4. The gait characteristic intelligent analysis method for spinal brace optimization according to claim 3 is characterized in that: The method of dividing the time series data into time windows of preset lengths based on the standardized gait data by a recursive neural network model, processing the continuous gait data by a sliding window technique, and updating the recursive neural network model parameters by a gradient descent algorithm comprises the following steps: S1231, initializing the weight and learning rate of the recursive neural network model based on the standardized gait data; S1232, using a sliding window technique, dividing the time series data into time windows of a preset length, reading gait data points from the standardized gait data stream, adding the gait data points to the time window, and extracting all gait data points within the time window; S1233, based on each gait data point in the time window, executing the forward propagation process of the recursive neural network model, and combining the back propagation algorithm to calculate the gradient of the loss function to the recursive neural network model parameters; S1234, combining the calculated gradient with a preset learning rate, and updating the parameters of the recursive neural network model using a gradient descent algorithm; S1235. Repeat steps S1232 to S1234 to read new gait data points from the gait data stream until the gait data stream processing is completed.

5. The gait characteristic intelligent analysis method for spinal brace optimization according to claim 4, characterized in that: The method of combining the dynamic principal component analysis algorithm with the multimodal principal component analysis algorithm to analyze the gait feature vector and obtain the gait feature parameters includes the following steps: S151, performing singular value decomposition on the gait feature vector to obtain an initial principal component matrix; S152, in each time window, using a dynamic principal component analysis algorithm and combining it with a recursive algorithm to update the principal component matrix, projecting the gait data through the updated principal component matrix, and extracting dynamic principal components; S153, based on the extracted dynamic principal components, using a multimodal principal component analysis algorithm, defining a joint optimization objective function, and optimizing the principal component matrix by an alternating minimization algorithm; S154. Use the optimized principal component matrix to extract features from the gait data to obtain gait feature parameters.

6. The gait characteristic intelligent analysis method for spinal brace optimization according to claim 5, characterized in that: The expression for optimizing the principal component matrix by the alternating minimization algorithm is: Where U represents the shared principal component matrix; M represents the number of modes; m represents the index value; X m represents the gait data matrix of the mth mode; V m represents the specific principal component matrix of the mth mode; T represents the transpose of the matrix; ||.|| F represents the Frobenius norm.

7. The gait characteristic intelligent analysis method for spinal brace optimization according to claim 1, characterized in that: The method of combining gait characteristic parameters with predefined brace characteristic parameters to construct a deep neural network model, inputting the real-time gait characteristic parameters into the deep neural network model to obtain predicted brace characteristic parameters, and optimizing the deep neural network model based on the brace characteristic parameters includes the following steps: S21, combining gait characteristic parameters with predefined brace characteristic parameters to establish a characteristic parameter set; S22. Based on the feature parameter set, a deep neural network model is constructed using a deep neural network algorithm, and the deep neural network model is trained using the gait feature parameters; S23, collecting real-time gait characteristic parameters, performing data processing on the real-time gait characteristic parameters, extracting real-time gait category parameters, and obtaining real-time gait characteristic parameters based on the real-time gait category parameters; S24, inputting the real-time gait characteristic parameters into the trained deep neural network model to obtain predicted brace characteristic parameters; S25. Based on the predicted brace characteristic parameters, use the controller to adjust the predefined brace characteristic parameters, collect feedback information, and optimize the deep neural network model based on the feedback information.

8. The gait characteristic intelligent analysis method for spinal brace optimization according to claim 7, characterized in that: The method of constructing a deep neural network model based on the feature parameter set using a deep neural network algorithm and training the deep neural network model using gait feature parameters includes the following steps: S221. Based on the feature parameter set, using a deep neural network algorithm, construct a deep neural network model including an input layer, a hidden layer, and an output layer, and establish a mapping relationship between gait feature parameters and brace feature parameters; S222, using the mean square error as the loss function of the deep neural network model, and randomly initializing the parameters of the deep neural network model; S223, inputting the gait characteristic parameters into the deep neural network model, obtaining the predicted brace characteristic parameters through the forward propagation algorithm, and calculating the loss value using the loss function according to the predicted brace characteristic parameters and the actual brace characteristic parameters; S224, based on the loss value, using the back propagation algorithm to calculate the gradient of each parameter, and update the parameters based on the gradient; S225. Repeat steps S223 to S224 until a preset number of iterations is reached.

9. The gait characteristic intelligent analysis method for spinal brace optimization according to claim 8, characterized in that: The expression of the controller is: Where u(t) represents the control output at time t; K p represents the proportional gain; e(t) represents the error between the predicted brace characteristic parameters and the actual brace characteristic parameters; K i represents the integral gain; t represents the proportion; τ represents the integral variable in the integral term; K d Represents the differential gain.

10. The gait characteristic intelligent analysis method for spinal brace optimization according to claim 1, characterized in that: The method of constructing a spine individual gait feature library based on gait feature parameters and brace feature parameters, collecting new gait data, and updating the spine individual gait feature library using the new gait feature data includes the following steps: S31, establishing a gait feature table based on the gait feature parameters; S32. Establishing a brace feature table based on the brace feature parameters; S33, constructing a spine individual gait feature library according to the gait feature table and the brace feature table, and in combination with the patient's feedback information; S34, using the stream processing platform to collect new gait data in real time, and setting the data collection frequency and data format based on the new gait data; S35, using a stream processing framework to perform data cleaning, denoising and standardization processing on the new gait data, and extracting new gait feature parameters based on the processed new gait data; S36, storing the new gait feature parameters into the spine individual gait feature library, and setting data sharding and replication strategies.

Images