Shoe and boot customization system based on model learning

Abstract

The invention relates to the technical field of shoe customization, in particular to a shoe customization system based on model learning. The system comprises a model friction unit, a weighted prediction unit, a bending crack unit and a customized boot unit. The crack prediction module predicts the bending points of the bottom of the shoe by using the shoe neural network model according to the average bending point of the surface of the shoe, the average bending degree of the bottom of the shoe and the acquired foot type data of the user, and connects the predicted bending points of the bottom of the shoe into a line by using a linear interpolation method. The shoe bottom crack probability is predicted according to the line and the obtained foot type data of the user, the shoe sole crack area is judged according to the predicted shoe bottom crack probability, the shoe bottom crack area can be predicted by analyzing the foot type data of the user, a more durable material can be used in a high crack risk area, and the user experience is improved. And lighter materials are used in the low-risk area, so that the shoe replacement frequency of a user is reduced, the maintenance cost of the shoe is reduced, and the overall performance of the shoe is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of shoe customization, and in particular to a shoe customization system based on model learning. Background Art

[0002] The shoe customization system uses advanced machine learning, computer vision, material science and optimization algorithm technologies to accurately identify areas with frequent friction on the soles of shoes and boots, areas with high damage on the surface of shoes and boots, and areas with cracks on the soles, and takes targeted reinforcement measures, including the application of high-wear-resistant materials, multi-layer structure design, selection of high-strength materials, embedded reinforcement layers, and the use of composite materials. When the user selects the size, shape, material and pattern of the customized shoes and boots and customizes them, the pressure distribution on the soles of the shoes and boots is different when the user wears the customized shoes and boots when walking, standing and running, as well as the bending degree of the soles and surfaces of the shoes and boots. The degree of friction is different, so the customized shoes and boots will have frequent friction areas on the bottom of the shoes and boots, high damage areas on the surface of the shoes and boots, and crack areas on the soles in a short period of time. When the customized shoe and boot model cannot identify and predict the frequent friction areas on the bottom of the shoes and boots, high damage areas on the surface of the shoes and boots, and the crack areas on the soles, and cannot use more durable materials in the high friction frequent areas, high damage areas, and high crack risk areas, the user's shoe replacement frequency will increase, which will lead to a decrease in the overall performance of the customized shoes and boots. At the same time, a large amount of lightweight and wear-resistant materials are used in low-risk areas, which increases the customization cost of shoes and boots. Therefore, we provide a shoe and boot customization system based on model learning. Summary of the invention

[0003] The purpose of the present invention is to provide a shoe customization system based on model learning to solve the problems raised in the above background technology.

[0004] To achieve the above-mentioned object, the present invention provides a shoe customization system based on model learning, comprising a model friction unit, a weighted prediction unit, a bending crack unit and a shoe customization unit;

[0005] The model friction unit acquires the user's foot shape data and shoe state data to establish a shoe neural network model, the shoe neural network model predicts the pressure value of the divided area according to the acquired user's foot shape data, and uses the predicted pressure value of the divided area to determine the frequent friction area on the bottom of the shoe;

[0006] The weighted prediction unit is used to receive the shoe state data and the shoe neural network model obtained in the model friction unit, input the obtained shoe state data into the shoe neural network model, the shoe neural network model calculates the pooling layer activation value according to the collected shoe surface bending degree data, and recalculates the pooling layer activation value and the full connection layer activation value, predicts the shoe surface bending point according to the recalculated pooling layer activation value and the full connection layer activation value, and determines the high damage area of ​​the shoe surface according to the predicted shoe surface bending point;

[0007] The bending crack unit is used to receive the shoe state data obtained in the model friction unit, the shoe neural network model and the shoe surface bending points predicted in the weighted prediction unit, and input the predicted shoe surface bending points into the shoe neural network model. The shoe neural network model calculates the average bending degree of the shoe bottom and the average bending points of the shoe surface according to the obtained shoe state data and the predicted shoe surface bending points, and then predicts the bending points of the shoe bottom and connects them into a line, and uses the line to predict the probability of cracks on the shoe bottom and determine the crack area of ​​the sole;

[0008] The customized shoe unit is used to receive the frequent friction areas on the bottom of the shoe in the model friction unit, the high damage areas on the surface of the shoe in the shoe neural network model and the weighted prediction unit, and receive the sole crack areas in the bending crack unit. The shoe neural network model takes strengthening measures for the frequent friction areas on the bottom of the shoe, the high damage areas on the surface of the shoe, and the sole crack areas during the process of customizing the user's shoes.

[0009] As a further improvement of the technical solution, the model friction unit includes a model building module and a friction area module;

[0010] The model building module obtains the user's foot shape data and shoe preference data, matches similar smart shoes and boots according to the obtained user's foot shape data, and then obtains the user's data in standing, walking, and running states through the smart shoes and boots, collects shoe bottom pressure data, shoe bottom and shoe surface bending degree data from the obtained shoe state data, and establishes a shoe neural network model according to the obtained user's foot shape data and the obtained user's shoe preference data;

[0011] The friction area module is used to receive the shoe bottom pressure data and the shoe neural network model collected in the model building module, input the collected shoe bottom pressure data into the shoe neural network model, and the shoe neural network model uses the acquired user foot shape data as the input feature vector , the shoe neural network model is based on the input feature vector Predict the pressure value of the divided area, and use the predicted pressure value of the divided area to judge the high pressure area. When the predicted pressure value of the divided area is greater than the set pressure value, the predicted divided area is determined to be a high pressure area, and the predicted divided area is an area with frequent friction on the bottom of the shoe.

[0012] As a further improvement of the technical solution, the principle of implementing the pressure value of the predicted divided area in the friction area module is as follows:

[0013] Through the shoe neural network model according to the input feature vector Predict the pressure value of the divided area and obtain the predicted pressure value of the divided area , the specific algorithm formula is:

[0014] ;

[0015] in, Refers to the The bottom of the shoe The predicted probability of a region ranges from between, Refers to the first User foot data, refers to Activation function, refers to the activation function, refers to the weight matrix, Refers to the bias vector.

[0016] As a further improvement of the technical solution, the weighted prediction unit includes a weighted activation module and a data calculation module;

[0017] The weighted activation module is used to receive the shoe surface curvature data and the shoe neural network model collected in the model building module, input the collected shoe surface curvature data into the shoe neural network model, and then use the collected shoe surface curvature data and the acquired user foot shape data as input feature vectors. , , using the convolutional neural network forward propagation formula according to the input feature vector Calculate the weighted sum of the convolution layer of the shoe neural network model, then use the calculated weighted sum of the convolution layer to calculate the activation value of the convolution layer of the shoe neural network model, calculate the weighted sum of the pooling layer of the shoe neural network model according to the calculated convolution layer activation value, and then calculate the activation value of the pooling layer of the shoe neural network model according to the calculated weighted sum of the pooling layer;

[0018] The data calculation module is used to receive the pooling layer activation value calculated in the weighted activation module to recalculate the weighted sum of the pooling layer of the shoe neural network model, and recalculate the activation value of the pooling layer of the shoe neural network model using the recalculated weighted sum of the pooling layer, and use the convolutional neural network forward propagation formula according to the input feature vector Calculate the weighted sum of the fully connected layer of the shoe neural network model, calculate the activation value of the fully connected layer of the shoe neural network model based on the calculated weighted sum of the fully connected layer, then calculate the weighted sum of the fully connected layer of the shoe neural network model again based on the calculated activation value of the fully connected layer, and use the recalculated weighted sum of the fully connected layer to calculate the activation value of the fully connected layer of the shoe neural network model again.

[0019] As a further improvement of the technical solution, the weighted prediction unit further includes a damage prediction module;

[0020] The damage prediction module is used to receive the pooling layer activation value and the fully connected layer activation value recalculated in the data calculation module, use the splicing layer in the shoe neural network model to splice the pooling layer activation value and the fully connected layer activation value recalculated, calculate the weighted sum of the splicing layer of the shoe neural network model according to the calculated splicing layer activation value, and then calculate the activation value of the splicing layer of the shoe neural network model again according to the calculated splicing layer weighted sum, add the weight matrices of the convolution layer, the pooling layer, the fully connected layer and the splicing layer, and then add the bias vectors of the convolution layer, the pooling layer, the fully connected layer and the splicing layer, predict the bending point of the shoe surface using the recalculated splicing layer activation value, the added weight matrix and the added bias vector, and determine the high damage area on the shoe surface according to the predicted bending point of the shoe surface.

[0021] As a further improvement of the technical solution, the bending crack unit includes an average bending module and a crack prediction module;

[0022] The average bending module is used to receive the shoe bottom bending degree data collected in the model building module, the shoe neural network model and the shoe surface bending point predicted in the damage prediction module, input the collected shoe bottom bending degree data and the predicted shoe surface bending point into the shoe neural network model and calculate the average bending degree of the shoe bottom and the average bending point of the shoe surface;

[0023] The crack prediction module is used for the average bending module to receive the average bending points of the shoe surface, the average bending degree of the shoe bottom and the shoe neural network model in the model building module. The shoe neural network model predicts the bending points of the shoe bottom according to the average bending points of the shoe surface, the average bending degree of the shoe bottom and the obtained user foot shape data, and uses the linear interpolation method to connect the predicted bending points of the shoe bottom into a line, and then uses the deep learning model to predict the probability of cracks in the shoe bottom according to the line and the obtained user foot shape data, and uses the predicted probability of cracks in the shoe bottom to determine whether the line is a sole crack. When the predicted probability of cracks in the shoe bottom is greater than the set probability of cracks in the shoe bottom, it is determined that the line is a sole crack, and the corresponding shoe bottom division area is found according to the sole crack, and the shoe bottom division area is the area where the sole crack exists.

[0024] As a further improvement of the present technical solution, the customized shoe unit is used to receive the frequent friction areas on the bottom of the shoe in the friction area module and the high damage areas on the surface of the shoe in the damage prediction module, receive the shoe neural network model in the model building module and the sole crack areas in the crack prediction module, and the shoe neural network model takes strengthening measures for the frequent friction areas on the bottom of the shoe, the high damage areas on the surface of the shoe, and the sole crack areas during the process of customizing the user's shoes.

[0025] Compared with the prior art, the present invention has the following beneficial effects:

[0026] 1. In the shoe customization system based on model learning, the crack prediction module uses the shoe neural network model to predict the bending points of the bottom of the shoe according to the average bending points of the shoe surface, the average bending degree of the bottom of the shoe and the obtained user foot shape data, and uses the linear interpolation method to connect the predicted bending points of the bottom of the shoe into a line. The probability of cracks in the bottom of the shoe is predicted according to the line and the obtained user foot shape data. The predicted probability of cracks in the bottom of the shoe is used to determine the crack area of ​​the sole. By analyzing the user's foot shape data, the area where cracks appear on the bottom of the shoe can be predicted. More durable materials can be used in high crack risk areas, and lighter materials can be used in low risk areas, which reduces the user's shoe replacement frequency and shoe repair costs, thereby improving the overall performance of the shoe.

[0027] 2. In the shoe customization system based on model learning, the damage prediction module uses the splicing layer in the shoe neural network model to splice the recalculated pooling layer activation value and the recalculated fully connected layer activation value, calculates the weighted sum of the splicing layer of the shoe neural network model according to the calculated splicing layer activation value and recalculates the activation value of the splicing layer of the shoe neural network model, adds the weight matrix and bias vector of the convolution layer, pooling layer, fully connected layer and splicing layer, and predicts the bending point of the shoe surface using the recalculated splicing layer activation value, the added weight matrix and the added bias vector, and determines the high damage area of ​​the shoe surface according to the predicted bending point of the shoe surface. By recalculating the splicing layer activation value, the weight matrix and the bias vector, the model can more accurately capture the stress distribution and bending characteristics of the shoe surface, and dynamically analyze the damage trend of the shoe surface under different usage conditions, thereby improving the accuracy of identifying high damage areas. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 A unit block diagram of the present invention;

[0029] Figure 2 It is a block diagram of the module unit of the present invention.

[0030] The meaning of each number in the figure is:

[0031] 1. Model friction unit; 11. Model building module; 12. Friction area module;

[0032] 2. Weighted prediction unit; 21. Weighted activation module; 22. Data calculation module; 23. Prediction damage module;

[0033] 3. Bending crack unit; 31. Average bending module; 32. Crack prediction module; 4. Customized footwear unit. DETAILED DESCRIPTION

[0034] The following will be combined with the accompanying drawings in 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.

[0035] Example 1

[0036] The present invention provides a shoe customization system based on model learning, please refer to Figure 1 , including a model friction unit 1, a weighted prediction unit 2, a bending crack unit 3 and a customized shoe unit 4;

[0037] The model friction unit 1 obtains the user's foot shape data and shoe state data to establish a shoe neural network model. The shoe neural network model predicts the pressure value of the divided area according to the obtained user's foot shape data, and uses the predicted pressure value of the divided area to judge the area with frequent friction on the bottom of the shoe; the weighted prediction unit 2 is used to receive the shoe state data and the shoe neural network model obtained in the model friction unit 1, and input the obtained shoe state data into the shoe neural network model. The shoe neural network model calculates the pooling layer activation value according to the collected shoe surface bending degree data, and recalculates the pooling layer activation value and the full connection layer activation value, predicts the bending point of the shoe surface according to the recalculated pooling layer activation value and the full connection layer activation value, and determines the high damage area of ​​the shoe surface according to the predicted shoe surface bending point; the bending crack unit 3 is used to receive the shoe state data obtained in the model friction unit 1 The shoe state data, the shoe neural network model and the predicted shoe surface bending points in the weighted prediction unit 2 are input into the shoe neural network model. The shoe neural network model calculates the average bending degree of the shoe bottom and the average bending point of the shoe surface according to the acquired shoe state data and the predicted shoe surface bending points, and then predicts the bending points of the shoe bottom and connects them into a line, and uses the line to predict the probability of cracks in the shoe bottom and judge the crack area of ​​the sole; the customized shoe unit 4 is used to receive the frequent friction area of ​​the shoe bottom in the model friction unit 1, the shoe neural network model and the high damage area of ​​the shoe surface in the weighted prediction unit 2, and receive the crack area of ​​the sole in the bending crack unit 3. The shoe neural network model takes strengthening measures for the frequent friction area of ​​the shoe bottom, the high damage area of ​​the shoe surface and the crack area of ​​the sole in the process of customizing the user's shoes.

[0038] The following is a refinement of the above units, see Figure 2 ;

[0039] The model friction unit 1 includes a model building module 11 and a friction area module 12;

[0040] The model building module 11 obtains the user's foot shape data and shoe preference data, and Match similar smart shoes and boots. After the user tries on the smart shoes and boots that are similar to the user's foot shape, the smart shoes and boots are used to obtain the user's data when standing, walking, and running. The pressure data of the bottom of the shoes and boots, the curvature data of the bottom of the shoes and boots and the surface of the shoes and boots are collected from the obtained shoe state data. Establish a shoe neural network model with the acquired user shoe preference data;

[0041] User foot data obtained Including the user's foot length, foot width, arch height, and toe shape data;

[0042] The user's footwear preference data obtained includes the user's preferred shoe type, pattern and material;

[0043] The friction area module 12 is used to receive the shoe bottom pressure data and the shoe neural network model collected in the model building module 11, and input the collected shoe bottom pressure data into the shoe neural network model. The shoe neural network model divides the shoe bottom into The shoe neural network model will obtain the user's foot shape data in the area The feature vector as input , the shoe neural network model is based on the input feature vector Predict the pressure value of the divided area, and use the predicted pressure value of the divided area and the set pressure value to judge the high-pressure area. When the predicted pressure value of the divided area is greater than the set pressure value, the predicted divided area is determined to be a high-pressure area, and the predicted divided area is marked as 1. The predicted divided area is the area with frequent friction on the bottom of the shoe. By predicting the pressure values ​​of different areas, it is possible to accurately identify which parts of the sole are subject to higher pressure. This area is often an area with frequent friction. Cushioning materials are added to the area with frequent friction to reduce impact and improve wearing comfort.

[0044] The principle of implementing the predicted pressure value of the divided area in the friction area module 12 is as follows:

[0045] Through the shoe neural network model according to the input feature vector Predict the pressure value of the divided area and obtain the predicted pressure value of the divided area , the specific algorithm formula is:

[0046] ;

[0047] in, Refers to the The bottom of the shoe The predicted probability of a region ranges from between, Refers to the first User foot data, refers to Activation function, Refers to the activation function. In neural networks, the activation function is used to introduce nonlinearity so that the model can fit more complex functional relationships. Refers to the weight matrix, the size of the matrix depends on the dimension of the input features and the number of neurons in the hidden layer. Refers to the bias vector. The size of the bias vector is the same as the number of neurons in the hidden layer. It is used to add a constant term to the input of each neuron. This formula is used to predict the pressure value of the divided area. By predicting the pressure values ​​of different areas, it is possible to accurately identify which parts of the sole are subject to higher pressure. Such areas are often areas with frequent friction. Adding cushioning materials to the areas with frequent friction can reduce impact force and improve wearing comfort.

[0048] The weighted prediction unit 2 includes a weighted activation module 21 and a data calculation module 22;

[0049] The weighted activation module 21 is used to receive the shoe surface curvature data collected in the model building module 11. and the shoe neural network model, which collects the shoe surface curvature data Input into the shoe neural network model, the shoe neural network model divides the shoe surface into area, and then collect the curvature data of the shoe surface And the user's foot data obtained The feature vector as input , , using the convolutional neural network forward propagation formula according to the input feature vector Calculate the weighted sum of the convolutional layer of the shoe neural network model, and then use the calculated weighted sum of the convolutional layer Calculate the activation value of the convolutional layer of the shoe neural network model and obtain the calculated convolutional layer activation value , according to the calculated convolutional layer activation value Calculate the weighted sum of the pooling layer of the shoe neural network model and obtain the calculated weighted sum of the pooling layer , where 2 refers to the pooling layer, and then the weighted sum of the calculated pooling layer The calculated pooling layer weighted sum calculates the activation value of the pooling layer of the shoe neural network model to obtain the calculated pooling layer activation value ,in, Refers to the activation function. In neural networks, the activation function is used to introduce nonlinearity so that the model can fit more complex functional relationships;

[0050] The weighted sum is calculated using the forward propagation formula of the convolutional neural network:

[0051] Collect input feature vectors Calculate the weighted sum of the convolutional layers of the shoe neural network model to obtain the calculated weighted sum of the convolutional layers , the specific algorithm formula is:

[0052] ;

[0053] in, Refers to the weighted sum of the convolutional layers calculated , 1 refers to the convolutional layer, Refers to the collected Data on the curvature of the shoe surface. Refers to the weight matrix of the convolutional layer in the shoe neural network model , Refers to the bias vector of the convolutional layer in the shoe neural network model , this formula is used to calculate the weighted sum of the convolutional layer of the shoe neural network model, which serves as the basis for the subsequent calculation of the activation value of the convolutional layer of the shoe neural network model;

[0054] The data calculation module 22 is used to receive the pooling layer activation value calculated in the weighted activation module 21 , based on the calculated pooling layer activation value Recalculate the weighted sum of the pooling layer of the shoe neural network model, and use the recalculated weighted sum of the pooling layer Calculate the activation value of the pooling layer of the shoe neural network model again to obtain the recalculated pooling layer activation value , using the convolutional neural network forward propagation formula according to the input feature vector Calculate the weighted sum of the fully connected layer of the shoe neural network model, according to the calculated weighted sum of the fully connected layer Calculate the activation value of the fully connected layer of the shoe neural network model and obtain the calculated activation value of the fully connected layer , using the calculated fully connected layer activation value Recalculate the weighted sum of the fully connected layer of the shoe neural network model, and use the recalculated weighted sum of the fully connected layer Calculate the activation value of the fully connected layer of the shoe neural network model again to obtain the recalculated fully connected layer activation value ;

[0055] Calculate the weighted sum of the pooling layer again to achieve the following:

[0056] Collect the calculated pooling layer activation values Calculate the weighted sum of the pooling layer of the shoe neural network model again to obtain the weighted sum of the pooling layer calculated again , the specific algorithm formula is:

[0057] ;

[0058] in, Refers to the shoe surface curvature data weight matrix of the fully connected layer in the shoe neural network model , Refers to the bias vector of the shoe surface curvature data in the fully connected layer of the shoe neural network model , this formula is used to recalculate the weighted sum of the pooling layer, which serves as the basis for the subsequent recalculation of the activation value of the pooling layer of the shoe neural network model;

[0059] The implementation principle of using the forward propagation of the convolutional neural network to calculate the weighted sum of the convolutional layer is as follows:

[0060] Collect input feature vectors Calculate the weighted sum of the fully connected layer of the shoe neural network model to obtain the calculated weighted sum of the fully connected layer , the specific algorithm formula is:

[0061] ;

[0062] in, Refers to the weighted sum of the convolutional layers calculated , 3 refers to the fully connected layer, Refers to the first User foot data, Refers to the user foot shape data weight matrix of the fully connected layer in the shoe neural network model , Refers to the user foot shape data bias vector of the fully connected layer in the shoe neural network model , this formula is used to calculate the weighted sum of the fully connected layer of the shoe neural network model, which serves as the basis for the subsequent calculation of the activation value of the fully connected layer of the shoe neural network model;

[0063] Calculate the weighted sum of the fully connected layer again to achieve the principle:

[0064] Collect the calculated fully connected layer activation values Calculate the weighted sum of the fully connected layer of the shoe neural network model again to obtain the weighted sum of the fully connected layer calculated again , the specific algorithm formula is:

[0065] ;

[0066] Among them, this formula is used to recalculate the weighted sum of the fully connected layer of the shoe neural network model, which serves as a basis for the subsequent recalculation of the activation value of the fully connected layer of the shoe neural network model;

[0067] The weighted prediction unit 2 also includes a damage prediction module 23;

[0068] The damage prediction module 23 is used to receive the pooling layer activation value calculated again in the data calculation module 22 And the fully connected layer activation value calculated again , using the concatenation layer in the shoe neural network model to recalculate the pooling layer activation value And the fully connected layer activation value calculated again Perform the splicing operation to obtain the calculated splicing layer activation value , where 4 refers to the concatenation layer, and the concatenation layer activation value is calculated based on the forward propagation formula of the convolutional neural network Calculate the weighted sum of the concatenated layers of the shoe neural network model, and calculate the weighted sum of the concatenated layers Calculate the activation value of the splicing layer of the shoe neural network model again to obtain the recalculated splicing layer activation value , add the weight matrices of the convolutional layer, pooling layer, fully connected layer, and splicing layer, then add the bias vectors of the convolutional layer, pooling layer, fully connected layer, and splicing layer, and use the recalculated splicing layer activation value , the added weight matrix and the added bias vector The bending points of the shoe surface are predicted. By recalculating the activation values, weight matrices and bias vectors of the splicing layer, the model can more accurately capture the stress distribution and bending characteristics of the shoe surface, and dynamically analyze the damage trend of the shoe surface under different usage conditions, thereby improving the accuracy of identifying high damage areas. According to the predicted bending points of the shoe surface, the corresponding shoe surface division area is found, and the corresponding shoe surface division area is determined to be a high damage area of ​​the shoe surface.

[0069] The implementation principle of using the forward propagation of the convolutional neural network to calculate the weighted sum of the splicing layer is as follows:

[0070] Collect the calculated concatenated layer activation values Calculate the weighted sum of the concatenated layers of the shoe neural network model to obtain the calculated weighted sum of the concatenated layers , the specific algorithm formula is:

[0071] ;

[0072] in, Refers to the weight matrix of the concatenation layer in the shoe neural network model , Refers to the bias vector of the concatenation layer in the shoe neural network model , this formula is used to calculate the weighted sum of the splicing layer of the shoe neural network model, which serves as the basis for the subsequent calculation of the activation value of the splicing layer of the shoe neural network model;

[0073] How the bending points of the shoe surface are predicted:

[0074] Collect the recalculated concatenated layer activation values , the added weight matrix and the added bias vector Predict the bending point of the shoe surface and obtain the predicted bending point of the shoe surface , the specific algorithm formula is:

[0075] ;

[0076] in, refers to Activation function: This formula is used to predict the bending point of the shoe surface. By recalculating the activation value of the splicing layer, the characteristics of the shoe surface can be captured more accurately, thereby improving the accuracy of the bending point prediction;

[0077] The bending crack unit 3 includes an average bending module 31 and a crack prediction module 32;

[0078] The average bending module 31 is used to receive the shoe bottom bending degree data collected in the model building module 11, the shoe neural network model and the shoe surface bending point predicted in the damage prediction module 23 , the collected shoe bottom bending degree data and the predicted shoe surface bending point The input is input into the shoe neural network model, and the shoe neural network model extracts the bending degree of the shoe bottom at the dth position under the cth load from the collected shoe bottom bending degree data. , and the number of loads on the bottom of the shoe , according to the curvature of the bottom of the shoe Load times Calculate the average curvature of the sole of the shoe By calculating the average bending degree, the durability of the sole material of the shoe under repeated bending can be quantified, so as to select more suitable materials. The shoe neural network model is used to obtain the collected shoe surface bending degree data. Number of loads on the surface of the shoe , based on the predicted bending points of the shoe surface and shoe surface load times Calculate the average bending point on the shoe surface , by calculating the average bending points, it is possible to determine where the surface of the shoe is repeatedly bent during use and is subject to higher stresses and loads;

[0079] The crack prediction module 32 is used to receive the average bending point of the shoe surface from the average bending module 31. , the average curvature of the sole of the shoe and establish a shoe neural network model in the model module 11, the shoe neural network model is based on the average bending point of the shoe surface , the average curvature of the sole of the shoe And the user's foot data obtained Predict the bending point of the sole of the shoe. By predicting the bending point of the sole, the unnatural bending of the sole during walking can be reduced, pressure and friction on the foot can be avoided, and the user's shoe comfort can be improved. The predicted bending point of the sole of the shoe can be output. , the predicted bending point of the shoe bottom Arrange in a certain order to form an orderly set of bending points on the bottom of the shoe , using linear interpolation to calculate the predicted bending point of the bottom of the shoe Connect into lines ,in, is the interpolation parameter, and The adjacent bending points of the bottom of the shoe are then identified using the deep learning model according to the line And the user's foot data obtained Predict the probability of cracks on the bottom of shoes and boots. By analyzing the user's foot shape data, the area where cracks will appear on the bottom of the shoes and boots can be predicted. More durable materials can be used in high-risk areas, and lighter materials can be used in low-risk areas. This reduces the user's shoe replacement frequency and shoe repair costs, improves the overall performance of the shoes, and obtains the predicted probability of cracks on the bottom of the shoes and boots. , using the predicted probability of cracks on the bottom of the shoe The probability of cracks on the bottom of the shoe is determined by the set Is it a crack in the sole? When the predicted probability of a crack in the sole of a shoe is When the probability of cracks on the bottom of the shoe is greater than the set probability, the judgment line The sole crack is a crack in the shoe. The corresponding shoe bottom division area is found according to the sole crack. The shoe bottom division area is the area where the sole crack exists.

[0080] The customized shoe unit 4 is used to receive the frequent friction areas on the bottom of the shoe in the friction area module 12 and the high damage areas on the surface of the shoe in the damage prediction module 23, receive the shoe neural network model in the model building module 11 and the sole crack areas in the crack prediction module 32. The shoe neural network model takes strengthening measures for the frequent friction areas on the bottom of the shoe, the high damage areas on the surface of the shoe and the sole crack areas during the process of customizing the user's shoes. By strengthening the vulnerable areas, the frequency of damage of the shoes can be reduced, the user's maintenance cost can be reduced, and the durability of the shoes can be improved. After the durability of the shoes is improved, the user can use the same pair of shoes for a longer time, reduce the frequency of replacement, and save expenses.

[0081] 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 shoe customization system based on model learning, characterized by: It includes a model friction unit (1), a weighted prediction unit (2), a bending crack unit (3) and a customized shoe unit (4); The model friction unit (1) acquires the user's foot shape data and shoe state data to establish a shoe neural network model, the shoe neural network model predicts the pressure value of the divided area according to the acquired user's foot shape data, and uses the predicted pressure value of the divided area to determine the area with frequent friction on the bottom of the shoe; The weighted prediction unit (2) is used to receive the shoe state data and the shoe neural network model obtained in the model friction unit (1), input the obtained shoe state data into the shoe neural network model, the shoe neural network model calculates the pooling layer activation value according to the collected shoe surface bending degree data, and recalculates the pooling layer activation value and the fully connected layer activation value, predicts the shoe surface bending point according to the recalculated pooling layer activation value and the fully connected layer activation value, and determines the high damage area of ​​the shoe surface according to the predicted shoe surface bending point; The bending crack unit (3) is used to receive the shoe state data obtained in the model friction unit (1), the shoe neural network model and the shoe surface bending points predicted in the weighted prediction unit (2), and input the predicted shoe surface bending points into the shoe neural network model. The shoe neural network model calculates the average bending degree of the shoe bottom and the average bending points of the shoe surface based on the obtained shoe state data and the predicted shoe surface bending points, and then predicts the bending points of the shoe bottom and connects them into a line, using the line to predict the probability of cracks in the shoe bottom and determine the crack area of ​​the shoe bottom; The customized shoe unit (4) is used to receive the frequent friction area on the bottom of the shoe in the model friction unit (1), the high damage area on the surface of the shoe in the shoe neural network model and weighted prediction unit (2), and receive the sole crack area in the bending crack unit (3). The shoe neural network model takes strengthening measures for the frequent friction area on the bottom of the shoe, the high damage area on the surface of the shoe and the sole crack area during the process of customizing the user's shoes.

2. The shoe customization system based on model learning according to claim 1, characterized in that: The model friction unit (1) comprises a model building module (11) and a friction area module (12); The model building module (11) obtains the user's foot shape data and shoe preference data, matches similar smart shoes and boots according to the obtained user's foot shape data, and then obtains the user's data in the standing, walking, and running states through the smart shoes and boots, collects shoe bottom pressure data, shoe bottom and shoe surface bending degree data from the obtained shoe state data, and establishes a shoe neural network model according to the obtained user's foot shape data and the obtained user's shoe preference data; The friction area module (12) is used to receive the shoe bottom pressure data and the shoe neural network model collected in the model building module (11), input the collected shoe bottom pressure data into the shoe neural network model, and the shoe neural network model uses the acquired user foot shape data as an input feature vector , the shoe neural network model is based on the input feature vector Predict the pressure value of the divided area, and use the predicted pressure value of the divided area to judge the high pressure area. When the predicted pressure value of the divided area is greater than the set pressure value, the predicted divided area is determined to be a high pressure area, and the predicted divided area is an area with frequent friction on the bottom of the shoe.

3. The shoe customization system based on model learning according to claim 2, characterized in that: The principle for implementing the prediction of the pressure value of the divided area in the friction area module (12) is as follows: Through the shoe neural network model according to the input feature vector Predict the pressure value of the divided area and obtain the predicted pressure value of the divided area , the specific algorithm formula is: ； in, Refers to the The bottom of the shoe The predicted probability of a region ranges from between, Refers to the first User foot data, refers to Activation function, refers to the activation function, refers to the weight matrix, Refers to the bias vector.

4. The shoe customization system based on model learning according to claim 2, characterized in that: The weighted prediction unit (2) comprises a weighted activation module (21) and a data calculation module (22); The weighted activation module (21) is used to receive the shoe surface curvature data and the shoe neural network model collected in the model building module (11), input the collected shoe surface curvature data into the shoe neural network model, and then use the collected shoe surface curvature data and the acquired user foot shape data as input feature vectors. , , using the convolutional neural network forward propagation formula according to the input feature vector Calculate the weighted sum of the convolution layer of the shoe neural network model, then use the calculated weighted sum of the convolution layer to calculate the activation value of the convolution layer of the shoe neural network model, calculate the weighted sum of the pooling layer of the shoe neural network model according to the calculated convolution layer activation value, and then calculate the activation value of the pooling layer of the shoe neural network model according to the calculated weighted sum of the pooling layer; The data calculation module (22) is used to receive the pooling layer activation value calculated in the weighted activation module (21) to recalculate the weighted sum of the pooling layer of the shoe neural network model, use the recalculated pooling layer weighted sum to recalculate the activation value of the pooling layer of the shoe neural network model, and use the convolutional neural network forward propagation formula to calculate the activation value of the pooling layer according to the input feature vector Calculate the weighted sum of the fully connected layer of the shoe neural network model, calculate the activation value of the fully connected layer of the shoe neural network model based on the calculated weighted sum of the fully connected layer, then calculate the weighted sum of the fully connected layer of the shoe neural network model again based on the calculated activation value of the fully connected layer, and use the recalculated weighted sum of the fully connected layer to calculate the activation value of the fully connected layer of the shoe neural network model again.

5. The shoe customization system based on model learning according to claim 4, characterized in that: The weighted prediction unit (2) further comprises a damage prediction module (23); The damage prediction module (23) is used to receive the pooling layer activation value and the fully connected layer activation value recalculated in the data calculation module (22), use the splicing layer in the shoe neural network model to splice the pooling layer activation value and the fully connected layer activation value recalculated, calculate the weighted sum of the splicing layer of the shoe neural network model according to the calculated splicing layer activation value, and then calculate the activation value of the splicing layer of the shoe neural network model again according to the calculated splicing layer weighted sum, add the weight matrices of the convolution layer, the pooling layer, the fully connected layer and the splicing layer, and then add the bias vectors of the convolution layer, the pooling layer, the fully connected layer and the splicing layer, use the recalculated splicing layer activation value, the added weight matrix and the added bias vector to predict the bending point of the shoe surface, and determine the high damage area of ​​the shoe surface according to the predicted bending point of the shoe surface.

6. The shoe customization system based on model learning according to claim 5, characterized in that: The curved crack unit (3) comprises an average curvature module (31) and a crack prediction module (32); The average bending module (31) is used to receive the shoe bottom bending degree data collected in the model building module (11), the shoe neural network model and the shoe surface bending point predicted in the damage prediction module (23), input the collected shoe bottom bending degree data and the predicted shoe surface bending point into the shoe neural network model and calculate the average bending degree of the shoe bottom and the average bending point of the shoe surface; The crack prediction module (32) is used to receive the average bending point of the shoe surface, the average bending degree of the shoe bottom and the shoe neural network model in the model building module (11) in the average bending module (31). The shoe neural network model predicts the bending point of the shoe bottom based on the average bending point of the shoe surface, the average bending degree of the shoe bottom and the acquired user foot shape data, connects the predicted bending points of the shoe bottom into a line using a linear interpolation method, and then uses a deep learning model to predict the probability of a crack in the shoe bottom based on the line and the acquired user foot shape data. The predicted probability of a crack in the shoe bottom is used to determine whether the line is a sole crack. When the predicted probability of a crack in the shoe bottom is greater than the set probability of a crack in the shoe bottom, the line is determined to be a sole crack. The corresponding shoe bottom division area is found based on the sole crack, and the shoe bottom division area is a sole crack area.

7. The shoe customization system based on model learning according to claim 2, characterized in that: The shoe customization unit (4) is used to receive the frequent friction area on the bottom of the shoe in the friction area module (12) and the high damage area on the surface of the shoe in the damage prediction module (23), receive the shoe neural network model in the model establishment module (11) and the sole crack area in the crack prediction module (32), and the shoe neural network model takes strengthening measures for the frequent friction area on the bottom of the shoe, the high damage area on the surface of the shoe and the sole crack area during the process of customizing the user's shoes.

Images