Intelligent shoe system for monitoring foot state
Through the combination of support vector machine, decision tree and neural network algorithm, the problem of insufficient classification accuracy in early detection of thumb valgus is solved, and higher detection accuracy and model generalization ability are achieved.
Patent Information
- Application Number
- CN202411759450.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-03
- Publication Date
- 2025-03-18
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
In early detection of thumb valgus, some machine learning algorithms are used alone, such as support vector machines, decision trees or neural networks, which have problems such as insufficient classification accuracy, overfitting and local optimal solutions, making it difficult to accurately distinguish between normal and early foot states of thumb valgus.
The combination method of support vector machine, decision tree and neural network algorithm is adopted, and it is executed in sequence and progressed layer by layer. Support vector machines are used for preliminary classification, decision tree optimization classification details, neural networks deeply explore complex relationships, and improve detection accuracy through the synergy of multiple algorithms.
Through the synergy of multiple algorithms, the accuracy of early detection of thumb valgus is improved, the generalization ability of the model is enhanced, and the limitations of a single algorithm are avoided.
Smart Images
Figure CN119650064A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of machine learning algorithm fusion, and in particular to an intelligent shoe system for monitoring foot status. Background Art
[0002] The fusion of machine learning algorithms is an important technology. With the development of science and technology, the application of machine learning algorithms in the medical field has gradually attracted attention. However, in the early detection of hallux valgus, the application of a single machine learning algorithm still has limitations.
[0003] For example, when only the support vector machine algorithm is used, if the data feature extraction is insufficient or the kernel function is improperly selected, the classification accuracy will be affected, and it will be difficult to accurately distinguish between normal and early hallux valgus foot conditions. The decision tree algorithm used alone may rely too much on the single division of data features, resulting in poor classification of complex foot data, overfitting, and reduced generalization ability of the model. Although the neural network algorithm has strong nonlinear fitting capabilities, the training process is complicated and it is easy to fall into local optimal solutions. Moreover, if the input feature selection is unreasonable, its advantages cannot be fully utilized. In order to solve this technical problem, we provide a smart shoe system for monitoring foot status. Summary of the invention
[0004] The purpose of the present invention is to provide an intelligent shoe system for monitoring foot status to solve the problems raised in the above background technology.
[0005] In the present invention, the support vector machine algorithm, the decision tree algorithm and the neural network algorithm are executed in sequence, the decision tree algorithm performs secondary classification on the data according to the classification result of the support vector machine algorithm, the neural network algorithm receives the data output by the decision tree algorithm for re-learning and classification, and each classification result is used as the input adjustment item of the next algorithm to perform early detection of hallux valgus.
[0006] To achieve the above-mentioned purpose, a smart shoe system for monitoring foot status is provided, comprising a data acquisition unit, an analysis and integration unit, and an abnormality judgment unit;
[0007] The data acquisition unit collects pressure data and joint angle data of the foot through pressure sensors arranged on the inner side of the forefoot of the sole, below the first metatarsophalangeal joint, and on the inner side of the heel, and an angle sensor located at the first metatarsophalangeal joint of the foot;
[0008] The analysis and integration unit establishes a three-dimensional biomechanical model of the foot, extracts key characteristic parameters from the collected foot pressure and angle data, uses a support vector machine algorithm, takes the characteristic parameters extracted from the foot data as input vectors, and inputs them into the three-dimensional biomechanical model of the foot. At the same time, the labeled normal and early hallux valgus patient data are used as training sets, and the two types of data are distinguished by finding the optimal classification hyperplane, and a decision tree algorithm is introduced to construct a tree classification structure according to the range of characteristic parameter values. Finally, a multi-layer structure is constructed by a neural network algorithm to learn the nonlinear relationship of the characteristic parameters and output the model classification results.
[0009] The abnormality judgment unit sets a threshold value for the difference in characteristic parameter distribution of the two groups of people according to the model classification result and based on statistics. During actual detection, the user's foot data is input into the three-dimensional biomechanical model of the foot, and the output characteristic parameters are compared with the threshold value. Based on the comparison result, it is judged whether there is an early abnormality of hallux valgus.
[0010] As a further improvement of the present technical solution, the analysis and integration unit includes a parameter extraction module, which extracts key characteristic parameters from the collected foot pressure and angle data, specifically including:
[0011] The product of the peak pressure of the first metatarsophalangeal joint and the duration and the rate of change of the big toe joint angle, where:
[0012] The peak pressure of the first metatarsophalangeal joint was obtained by analyzing the foot pressure data using a peak detection algorithm, and the duration was recorded by a timestamp;
[0013] The rate of change of the big toe joint angle is obtained by differential calculation based on the angle sensor data.
[0014] As a further improvement of the technical solution, the analysis and integration unit includes a model training module, and the support vector machine algorithm in the model training module uses a Gaussian radial basis kernel function for operation, as follows:
[0015] The sample data is marked, the normal foot sample is marked as -1, and the early-stage hallux valgus foot sample is marked as 1, wherein the characteristic parameters are included in the sample data;
[0016] The Gaussian radial basis kernel function is expressed as K(x i , x j )=exp(-γx i -x j 2 ), where x i 、x j is a vector containing four characteristic parameters, γ is a kernel function parameter, and the initial value of the kernel function parameter γ is determined according to the characteristic dimension of the training set data;
[0017] The labeled sample data is input into the support vector machine algorithm for training to obtain the optimal hyperplane. For the newly acquired foot data, the four feature parameters are recalculated, and the vector composed of the four feature parameters of the new data is input into the trained support vector machine model. The model calculates the category prediction result of whether the new data belongs to a normal foot or an early hallux valgus foot based on the optimal hyperplane obtained by previous training;
[0018] The prediction results are transmitted to the three-dimensional biomechanical model of the foot together with the four characteristic parameters of the new data as input.
[0019] As a further improvement of the technical solution, the optimal hyperplane is specifically explained as follows:
[0020] The optimal hyperplane equation can be expressed as ω T x+b=0; where x is the input feature vector, ω is the weight vector, and b is the bias term. ω and b are used to determine the position and direction of the hyperplane in the feature space. The direction of ω determines the normal direction of the hyperplane. b is used to translate the hyperplane and divide data into different categories. During the training process, the support vector machine uses an optimization algorithm to find the optimal ω. * and b * , to classify normal feet and feet in the early stages of hallux valgus.
[0021] As a further improvement of the technical solution, the analysis and integration unit includes a classification optimization module. When the decision tree algorithm is introduced into the model, the classification optimization module first uses the classification result of the support vector machine algorithm as the initial input feature of the decision tree algorithm, as follows:
[0022] The three-dimensional biomechanical model of the foot determines the category according to the position of the hyperplane and the position of the sample point relative to the hyperplane. If the sample point is on one side of the hyperplane, it is determined to be a normal category and outputs -1. If the sample point is on the other side of the hyperplane, it is determined to be an early hallux valgus category and outputs 1.
[0023] The classification results of the support vector machine algorithm are used as the initial input features of the decision tree algorithm, and are used as new feature dimensions in the form of one-hot encoding, together with the original four feature parameters, to construct a decision tree.
[0024] As a further improvement of the technical solution, the method of the classification optimization module in constructing the decision tree structure is as follows:
[0025] Calculate the information gain ratio of all feature parameters including the output of the support vector machine, and select the feature parameter with the largest information gain ratio as the root node. The information gain ratio is an indicator used to evaluate the importance of features in classification tasks. The root node is the starting point of the decision tree structure and is used to determine the first division criterion of the entire tree.
[0026] Determine the value range of the characteristic parameter according to medical standards and historical data, each value range corresponds to a branch, then recursively recalculate the information gain ratio of the remaining characteristic parameters for the sample subset corresponding to each branch, select the next best characteristic parameter as a child node and divide its value range to construct a subtree;
[0027] When all samples in the subset belong to the same category, the construction is stopped, and a tree classification structure is constructed based on the range of feature parameter values.
[0028] As a further improvement of the technical solution, the analysis and integration unit includes a neural network module. The neural network constructed by the neural network module uses the output of the decision tree algorithm and the category label of the support vector machine as part of the neuron input of the input layer, and inputs the above four characteristic parameters into the input layer at the same time, wherein the number of neurons in the input layer is determined according to the dimension of the input data, and the neural network is set to a three-layer structure, including an input layer, a hidden layer and an output layer, and the number of neurons in the hidden layer is equal to the number of neurons in the input layer. The number of neurons in the output layer is 1, which is used to output the early detection results of hallux valgus.
[0029] As a further improvement of the technical solution, the method for training the neural network module to learn the nonlinear relationship of the characteristic parameters is as follows:
[0030] For each sample in the training set, which includes the decision tree output, the support vector machine result, and the four feature parameters, a linear normalization method is used for normalization;
[0031] Randomly initialize the weight matrix W from the input layer to the hidden layer 1 and the weight matrix W from the hidden layer to the output layer 2 , and the hidden layer bias vector b 1 and the output layer bias vector b 2 ;
[0032] For the i-th normalized sample in the training set Input it to the input layer of the neural network, and in the process of propagation from the input layer to the hidden layer, calculate the input of the hidden layer neurons Where W 1 T It is W 1The transposed matrix of the hidden layer neurons is then used to apply the activation function to the input of the hidden layer neurons to obtain the output of the hidden layer neurons.
[0033] Calculate the input of the output layer neurons based on the output of the hidden layer neurons And because it is a binary classification problem, the output layer uses the Sigmoid activation function to get the predicted output represents the probability that the i-th sample is predicted to be in the early stage of hallux valgus;
[0034] For the m samples in the training set, the cross entropy loss function is used to measure the difference between the predicted results and the true labels, and the Adam gradient descent algorithm is used for back propagation and weight update, so that the neural network learns the nonlinear relationship between the decision tree output, the support vector machine results and the original foot feature parameters, and outputs the final classification result of the model.
[0036] For the N samples in the training set, the cross entropy loss function is used to measure the difference between the predicted results and the true labels, and the Adam gradient descent algorithm is used for back propagation and weight update, so that the neural network learns the nonlinear relationship between the decision tree output, the support vector machine results and the original foot feature parameters, and outputs the final classification result of the model.
[0037] As a further improvement of the technical solution, the abnormality judgment unit judges the early abnormality of hallux valgus according to the final classification result of the model, as follows:
[0038] When the trained neural network detects new foot data, the new data is processed in the same way and the output value is calculated through forward propagation. When the output value of the output layer is greater than the preset threshold, it is judged as early hallux valgus, and when it is less than or equal to the preset threshold, it is judged as normal.
[0039] Compared with the prior art, the present invention has the following beneficial effects:
[0040] In a smart shoe system for monitoring foot status, the support vector machine algorithm uses a Gaussian radial basis kernel function to optimize the initial value of the kernel function parameter according to the feature dimension of the training set, accurately determine the hyperplane, and lay the foundation for classification. The decision tree algorithm is guided by the classification result of the support vector machine, calculates the information gain ratio to build a tree structure, selects the best feature parameters to refine the classification, and makes up for the deficiency of single division. The neural network algorithm uses the output of the decision tree and the support vector machine as part of the input, constructs a multi-layer structure, reasonably sets the network parameters, and learns the nonlinear relationship between the feature parameters. The three algorithms are progressive layer by layer. The support vector machine provides preliminary classification, the decision tree optimizes the classification details, and the neural network deeply mines complex relationships. The synergistic effect improves the accuracy of early detection of hallux valgus. BRIEF DESCRIPTION OF THE DRAWINGS
[0041] Figure 1 It is an overall block diagram of the present invention.
[0042] The meaning of each number in the figure is:
[0043] 1. Data acquisition unit; 2. Analysis and integration unit; 21. Parameter extraction module; 22. Model training module; 23. Classification optimization module; 24. Neural network module; 3. Abnormal judgment unit. DETAILED DESCRIPTION
[0044] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0045] The present invention provides a smart shoe system for monitoring foot status. Figure 1 As shown, it includes a data acquisition unit 1, an analysis and integration unit 2 and an abnormality judgment unit 3;
[0046] The data acquisition unit 1 collects pressure data of the foot through pressure sensors arranged on the inner side of the forefoot of the sole, below the first metatarsophalangeal joint, and on the inner side of the heel, and accurately captures the changes in pressure distribution of the foot in various states such as walking and standing, which is crucial for early detection of pressure abnormalities caused by hallux valgus.
[0047] The angle sensor arranged at the first metatarsophalangeal joint of the foot collects the foot joint angle data and tracks the angle changes of the toe joint in three-dimensional space in real time.
[0048] When a user puts on the smart shoes equipped with this data acquisition unit 1 and starts activities, the pressure sensor and the angle sensor start working at the same time. The pressure sensor monitors the pressure of various parts of the sole in real time at a frequency of once per millisecond, and converts the pressure changes into electrical signals. These electrical signals are converted into digital signals by the built-in high-precision analog-to-digital converter, and then transmitted to the analysis and integration unit 2 through the data bus. Through comprehensive and accurate data acquisition, the characteristic changes related to the early stage of hallux valgus can be more accurately identified, thereby providing more accurate input for subsequent analysis algorithms.
[0049] The analysis and integration unit 2 establishes a three-dimensional biomechanical model of the foot, and the parameter extraction module 21 extracts key characteristic parameters from the collected foot pressure and angle data, as follows:
[0050] The pressure sensor continuously collects pressure data below the first metatarsophalangeal joint. Let the collected pressure data sequence be P(t), where t represents time. First, an initial threshold T is set. 1 , traverse the pressure data sequence P(t), when P(t)>T 1 When P(t)>P(t-1) and P(t)>P(t+1), it is initially judged as a possible peak point P peak Then, a secondary verification is performed near the peak point, i.e., in the interval [t-Δt, t+Δt], and the average value of the pressure data in this interval is calculated. like Then determine P peak is the actual pressure peak.
[0051] For each determined pressure peak, the first time it exceeds the threshold value T is recorded. 1 Time t start and finally falls below the threshold T 1 Time t end , calculate the duration D = t end -t start Finally, the product of the peak pressure and duration of the first metatarsophalangeal joint is obtained: 1 =P peak ×D.
[0052] The angle sensor collects the angle data sequence θ(t) of the first metatarsophalangeal joint of the foot. For two adjacent angle values θ(t) and θ(t+Δt) in the time series, Δt is the sampling time interval, and the angle change rate is The final change rate of the big toe joint angle F is 2 =R θ ,These feature parameters will be used as input of the subsequent support vector machine algorithm, ,decision tree algorithm and neural network algorithm for further foot status analysis and ,early detection of hallux valgus. Through this precise feature parameter extraction method, ,information closely related to foot health conditions can be mined from the original data, ,providing strong support for the smart shoe system to accurately judge the foot status.
[0053] The analysis and integration unit 2 includes a model training module 22. The model training module 22 uses the characteristic parameters extracted from the foot data as input vectors and inputs them into the three-dimensional biomechanical model of the foot. At the same time, the labeled normal and early hallux valgus patient data are used as training sets, and then the support vector machine algorithm is used to find the optimal classification hyperplane to distinguish the two types of data, as follows:
[0054] F 1 、F 2 The eigenvector F = (F 1 , F 2), and mark the samples into two categories, normal foot samples are marked as y = -1, and early hallux valgus foot samples are marked as y = 1, so that a labeled training set is obtained. Where n is the number of training samples.
[0055] For linearly separable data, the hyperplane equation can be expressed as ω T x+b=0,where ω=(ω 1 ,ω 2 ) is the weight vector and b is the bias term.
[0056] In order to process nonlinearly separable data, the Gaussian radial basis kernel function K(x i , x j ) = exp(-γ||x i -x j || 2 ), where γ is the kernel function parameter, ||x i -x j || 2 Represents two eigenvectors x i 、x j The kernel function is used to map the original feature space to a higher-dimensional feature space so that the data may become linearly separable in the new space.
[0057] The initial value of the kernel function parameter γ is determined according to the characteristic dimension of the training set data, and the optimization goal of the support vector machine is to minimize the following objective function sty (i) (ω T φ(x (i) )+b)≥1-ξ i ,ξ i ≥0, i=1,...,n; where φ(x) is the function that maps data to the high-order space, C is the penalty term used to control the degree of penalty for misclassified samples, ξ i is a slack variable used to handle noise and outliers in the data.
[0058] Use the optimization algorithm to get the optimal weight vector ω * and the bias term b * For the newly acquired foot data, the feature vector F composed of its dual feature parameters is calculated in the same way new =(F new1 , F new2 ), F new Input into the trained support vector machine model, and calculate the decision function f(F new )=ω *T φ(x new )+b *The value of is used to determine whether the new data belongs to a normal foot or an early-stage hallux valgus foot. If f(x new )>0, the prediction is early hallux valgus foot, and the output category label is 1. new )≤0, it is predicted to be a normal foot and the output category label is -1.
[0059] The category prediction results and feature vector x output by the support vector machine algorithm new The information is transmitted as input to the three-dimensional biomechanical model of the foot. The three-dimensional biomechanical model of the foot uses this information, combined with its internal modeling of the foot structure and mechanical principles, to further analyze and evaluate the state of the foot, providing a richer information basis for subsequent decision tree algorithms and neural network algorithms, and jointly achieving more accurate judgment of the foot state and effective detection of hallux valgus in the early stage.
[0060] When the classification optimization module 23 introduces the decision tree algorithm into the model, it first uses the classification result of the support vector machine algorithm as the initial input feature of the decision tree algorithm to construct a decision tree structure, as follows:
[0061] The support vector machine classification result of each sample is recorded as n is the number of training samples. The support vector machine classification results are one-hot encoded. Since the classification results are only -1 and 1, a two-dimensional vector is created to represent the one-hot encoding. For example, when When , the one-hot encoding is [1, 0]; when , the one-hot encoding is [0, 1]; this converts the support vector machine classification result into a new feature dimension, and each sample now has four features.
[0062] For the four features including the support vector machine classification result after one-hot encoding, the information gain ratio of each feature is calculated as follows:
[0063] Calculate the entropy of the dataset D where p k is the proportion of samples in the dataset that belong to the normal or early hallux valgus category.
[0064] For each feature F j (j=1, 2, 3, 4), calculate its conditional entropy Where m is the feature F j The number of values, p v It is feature F j The proportion of samples with value v, D v It is feature F j The sample subset whose value is v.
[0065] Then calculate the information gain g(D, F j )=H(D)-H(D|Fj ); Finally, calculate the information gain ratio in It is feature F j The entropy of .
[0066] Select the feature with the largest information gain ratio as the root node, and determine the root node feature F based on medical standards and historical data i For each sample subset corresponding to the value range, recursively recalculate the information gain ratio of the remaining features, select the next best feature as the child node and divide its value range to build a subtree, and repeat this process until the conditions are met.
[0067] After the decision tree is built, it can start from the root node according to the feature value of the new input sample, select the corresponding branch according to the value range of the feature value, and gradually judge downward, and finally classify the sample into a specific category. The tree structure of the decision tree can intuitively display the decision relationship between features, providing an easy-to-understand and explain model for the classification of foot status.
[0068] At the same time, it is combined with the support vector machine algorithm to further refine the classification process and improve the accuracy of the entire model in early detection of hallux valgus. In the subsequent fusion with the neural network algorithm, the output of the decision tree can be used as the input of the neural network, together providing a more comprehensive and accurate basis for the final judgment of the foot status.
[0069] Finally, a multi-layer structure is constructed through a neural network algorithm to learn the nonlinear relationship of feature parameters and output the model classification results, as follows:
[0070] Assume that the number of samples in the training set is m. For the i-th sample, it contains the decision tree output, the support vector machine result and two feature parameters. Each eigenvalue is mapped to the interval [0,1]. For feature j, it is normalized. After normalization, the eigenvalues of all samples are in the same numerical range, which is convenient for the training of the neural network.
[0071] Since the input data includes the decision tree output, the support vector machine result and two feature parameters, a total of four dimensions, n = 6, the neural network is set to a three-layer structure, and the number of hidden layer neurons is set to The number of neurons in the output layer is 1, which is used to output the early detection results of hallux valgus.
[0072] Randomly initialize the weight matrix W from the input layer to the hidden layer 1 and the weight matrix W from the hidden layer to the output layer 2 , and the hidden layer bias vector b 1 and the output layer bias vector b 2 , for the i-th normalized sample Input it to the input layer of the neural network and calculate the input of the hidden layer neurons Here W 1 T It is W 1 The transposed matrix of the hidden layer neurons is then used to apply the activation function to the input of the hidden layer neurons to obtain the output a of the hidden layer neurons. 1i , calculate the input of the output layer neurons Since it is a binary classification problem, the output layer uses the Sigmoid activation function to get the predicted output It represents the probability that the i-th sample is predicted to be in the early stage of hallux valgus.
[0073] For the m samples in the training set, the cross entropy loss function is used to measure the difference between the predicted results and the true labels, as follows:
[0074] Let the true label of the i-th sample be y i , then the calculation formula of the cross entropy loss function J is:
[0075]
[0076] Calculate the gradient of the output layer with respect to the loss function And the hidden layer to output layer weight W 2 and bias b 2 Gradient And calculate the gradient of the hidden layer with respect to the loss function in Expressed as element-wise multiplication.
[0077] Calculate the weight W from the input layer to the hidden layer 1 and bias b 1 Gradient The parameters of the Adam algorithm are initialized. In each training iteration, the first-order moment estimate and the second-order moment estimate are updated and corrected. Finally, the weights and biases of the neural network are updated according to the corrected results.
[0078] By iterating the above process multiple times, the weights and biases of the neural network are continuously adjusted so that the neural network can learn the nonlinear relationship between the decision tree output, the support vector machine results and the original foot feature parameters, thereby gradually optimizing the model and improving the accuracy of early detection of hallux valgus.
[0079] When the preset number of training rounds is met, the training ends and a trained neural network model is obtained, which is used to perform early detection of hallux valgus on new foot data and output the final classification result, and the classification result is output to the abnormality judgment unit 3 for abnormality judgment.
[0080] The abnormality judgment unit 3 collects a large amount of sample data marked as normal and early hallux valgus through the data collection unit 1, calculates the characteristic parameters of each sample through the analysis and integration unit, and performs statistical analysis on each characteristic parameter of the normal population and the early hallux valgus population, as follows:
[0081] Calculate the mean μ of each feature adoption number in the normal population n , standard deviation σ n and the mean μ in the early population of horse valgus a , standard deviation σ a , for a certain characteristic parameter, set the upper threshold T 1 =μ n +kσ n and the lower threshold T 2 =μ n -kσ n ; where k is a coefficient determined based on experience, and a threshold is set for each feature parameter according to the above method to obtain a set of threshold vectors T = [T 1 , T 2 , ..., T m ].
[0082] When the user wears the smart shoes for daily activities, the data acquisition unit 1 collects foot pressure and angle data in real time and transmits it to the analysis and integration unit 2. The analysis and integration unit 2 extracts characteristic parameters according to the previous method and then inputs these characteristic parameters into the trained three-dimensional biomechanical model of the foot, as follows:
[0083] The foot three-dimensional biomechanical model outputs the characteristic parameter value of the user's foot P = [P 1 , P 2 , ..., P m ], the abnormality judgment unit compares the output feature parameter P with the set threshold vector T. For each feature parameter P i , if P i >T i , then the characteristic parameter indicates that there is an abnormality, and the number of abnormal characteristic parameters is recorded. According to the abnormal characteristic parameters, it is comprehensively judged whether there is an early abnormality of hallux valgus, so as to subsequently track and analyze the changes in the user's foot health status.
[0084] In the present invention, the data acquisition unit collects pressure and angle data through sensors at the soles and foot joints, the analysis and integration unit establishes a three-dimensional biomechanical model of the foot, the parameter extraction module first extracts key feature parameters, and then the model training module is trained using a support vector machine algorithm, the classification optimization module introduces a decision tree algorithm optimization, the neural network module constructs a neural network to learn nonlinear relationships, the abnormality judgment unit sets a threshold based on statistics, and compares the model output feature parameters with the threshold to determine whether there is an early hallux valgus abnormality. By combining the advantages of multiple algorithms, the accuracy of early hallux valgus detection is improved, and intelligent monitoring and evaluation of the foot status is achieved.
[0085] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments. The above embodiments and descriptions are only preferred examples of the present invention and are not intended to limit the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention. The scope of protection of the present invention is defined by the attached claims and their equivalents.
Claims
1. A smart shoe system for monitoring foot status, characterized in that: It comprises a data collection unit (1), an analysis and integration unit (2) and an abnormality judgment unit (3); The data collection unit (1) collects pressure data and joint angle data of the foot through pressure sensors arranged on the inner side of the forefoot of the sole, below the first metatarsophalangeal joint and on the inner side of the heel, and an angle sensor located at the first metatarsophalangeal joint of the foot; The analysis and integration unit (2) establishes a three-dimensional biomechanical model of the foot, extracts key characteristic parameters from the collected foot pressure and angle data, and then uses a support vector machine algorithm to take the key characteristic parameters extracted from the foot data as input vectors and input them into the three-dimensional biomechanical model of the foot. At the same time, the labeled normal and early hallux valgus patient data are used as a training set, and the two types of data are distinguished by finding the optimal classification hyperplane in the support vector machine algorithm, and a decision tree algorithm is introduced to construct a tree classification structure according to the value range of the key characteristic parameters. Finally, a multi-layer structure is constructed by a neural network algorithm to learn the nonlinear relationship of the characteristic parameters, and the model classification result is output; The abnormality judgment unit (3) sets a threshold value for the difference in characteristic parameter distribution of the two groups of people according to the model classification result and based on statistics. During actual detection, the user's foot data is input into the three-dimensional biomechanical model of the foot, and the output characteristic parameters are compared with the threshold value. Based on the comparison result, it is judged whether there is an early hallux valgus abnormality.
2. The smart shoe system for monitoring foot status according to claim 1, characterized in that: The analysis and integration unit (2) comprises a parameter extraction module (21), wherein the parameter extraction module (21) extracts key characteristic parameters from the collected foot pressure and angle data, specifically comprising: The product of the peak pressure of the first metatarsophalangeal joint and the duration and the rate of change of the big toe joint angle, where: The peak pressure of the first metatarsophalangeal joint was obtained by analyzing the foot pressure data using a peak detection algorithm, and the duration was recorded by a timestamp; The rate of change of the big toe joint angle is obtained by differential calculation based on the angle sensor data.
3. The smart shoe system for monitoring foot status according to claim 1, characterized in that: The analysis and integration unit (2) includes a model training module (22), and the support vector machine algorithm in the model training module (22) uses a Gaussian radial basis kernel function for operation, as follows: The sample data is marked, the normal foot sample is marked as -1, and the early-stage hallux valgus foot sample is marked as 1, wherein the characteristic parameters are included in the sample data; The Gaussian radial basis kernel function is expressed as K(x i , x j )=exp(-γx i -x j 2 ); where x i 、x j is a vector containing four characteristic parameters, γ is a kernel function parameter, and the initial value of the kernel function parameter γ is determined according to the characteristic dimension of the training set data; The labeled sample data is input into the support vector machine algorithm for training to obtain the optimal hyperplane. For the newly acquired foot data, the four feature parameters are recalculated, and the vector composed of the four feature parameters of the new data is input into the trained support vector machine model. The model calculates the category prediction result of whether the new data belongs to a normal foot or an early hallux valgus foot based on the optimal hyperplane obtained by previous training; The prediction results are transmitted to the three-dimensional biomechanical model of the foot together with the four characteristic parameters of the new data as input.
4. The smart shoe system for monitoring foot status according to claim 3, characterized in that: The optimal hyperplane is specifically explained as follows: The optimal hyperplane equation can be expressed as ω T x+b=0; where x is the input feature vector, ω is the weight vector, and b is the bias term. ω and b are used to determine the position and direction of the hyperplane in the feature space. The direction of ω determines the normal direction of the hyperplane. b is used to translate the hyperplane and divide data into different categories. During the training process, the support vector machine uses an optimization algorithm to find the optimal ω. * and b * , to classify normal feet and feet in the early stages of hallux valgus.
5. The smart shoe system for monitoring foot status according to claim 1, characterized in that: The analysis and integration unit (2) includes a classification optimization module (23). When the decision tree algorithm is introduced into the model, the classification optimization module (23) first uses the classification result of the support vector machine algorithm as the initial input feature of the decision tree algorithm, as follows: The three-dimensional biomechanical model of the foot determines the category according to the position of the hyperplane and the position of the sample point relative to the hyperplane. If the sample point is on one side of the hyperplane, it is determined to be a normal category and outputs -1. If the sample point is on the other side of the hyperplane, it is determined to be an early hallux valgus category and outputs 1. The classification results of the support vector machine algorithm are used as the initial input features of the decision tree algorithm, and are used as new feature dimensions in the form of one-hot encoding, together with the four feature parameters, to construct a decision tree.
6. The smart shoe system for monitoring foot status according to claim 5, characterized in that: The method of the classification optimization module (23) in constructing the decision tree structure is as follows: Calculate the information gain ratio of all feature parameters including the output of the support vector machine, and select the feature parameter with the largest information gain ratio as the root node. The information gain ratio is an indicator used to evaluate the importance of features in classification tasks. The root node is the starting point of the decision tree structure and is used to determine the first division criterion of the entire tree. Determine the value range of the characteristic parameter according to medical standards and historical data, each value range corresponds to a branch, then recursively recalculate the information gain ratio of the remaining characteristic parameters for the sample subset corresponding to each branch, select the next best characteristic parameter as a child node and divide its value range to construct a subtree; When all samples in the subset belong to the same category, the construction is stopped, and a tree classification structure is constructed based on the range of feature parameter values.
7. The smart shoe system for monitoring foot status according to claim 1, characterized in that: The analysis and integration unit (2) comprises a neural network module (24), wherein the neural network constructed by the neural network module (24) uses the output of the decision tree algorithm and the category label of the support vector machine as the neuron input of the input layer, and simultaneously inputs the above four characteristic parameters into the input layer, wherein the number of neurons in the input layer is determined according to the dimension of the input data, and the neural network is set to a three-layer structure, including an input layer, a hidden layer and an output layer, and the number of neurons in the hidden layer is equal to the number of neurons in the input layer. The number of neurons in the output layer is 1, which is used to output the early detection results of hallux valgus.
8. The smart shoe system for monitoring foot status according to claim 7, characterized in that: The method for training the neural network module (24) to learn the nonlinear relationship of the characteristic parameters is as follows: For each sample in the training set, which includes the decision tree output, the support vector machine result, and the four feature parameters, a linear normalization method is used for normalization; Randomly initialize the weight matrix W1 from the input layer to the hidden layer and the weight matrix W2 from the hidden layer to the output layer, as well as the hidden layer bias vector b1 and the output layer bias vector b2; For the i-th normalized sample in the training set Input it to the input layer of the neural network, and in the process of propagation from the input layer to the hidden layer, calculate the input of the hidden layer neurons in is the transposed matrix of W1, and then the activation function is applied to the input of the hidden layer neurons to obtain the output of the hidden layer neurons Calculate the input of the output layer neurons based on the output of the hidden layer neurons And because it is a binary classification problem, the output layer uses the Sigmoid activation function to get the predicted output represents the probability that the i-th sample is predicted to be in the early stage of hallux valgus; For the m samples in the training set, the cross entropy loss function is used to measure the difference between the predicted results and the true labels, and the Adam gradient descent algorithm is used for back propagation and weight update, so that the neural network learns the nonlinear relationship between the decision tree output, the support vector machine results and the original foot feature parameters, and outputs the final classification result of the model.
9. The smart shoe system for monitoring foot status according to claim 1, characterized in that: The abnormality judgment unit (3) judges the early abnormality of hallux valgus according to the final classification result of the model, as follows: When the trained neural network detects new foot data, the new data is processed in the same way and the output value is calculated through forward propagation. When the output value of the output layer is greater than the preset threshold, it is judged as early hallux valgus, and when it is less than or equal to the preset threshold, it is judged as normal.