Shoe tree local fine tuning method and system based on deep learning

Abstract

The invention discloses a shoe tree local fine tuning method and system based on deep learning, and relates to the technical field of monitoring analysis, and the method comprises the steps: obtaining foot dynamic data of a user under various simulated preset motion scenes, extracting a biomechanical feature vector under each preset motion scene based on the foot dynamic data, and obtaining a biomechanical feature vector under each preset motion scene based on the biomechanical feature vector; analyzing the biomechanical feature vectors, generating function demand portraits corresponding to different motion modes, calculating a function demand difference degree between the different motion modes, comparing the function demand difference degree with a preset conflict threshold value, and identifying a local area with a function demand conflict on the shoe tree; and according to the identified local area with the function demand conflict, generating a shoe tree modular partition scheme and a corresponding adjustable module parameter set. The method has the effect of improving the local fine adjustment efficiency of the shoe tree.

Description

Technical Field

[0001] This application relates to the field of monitoring and analysis technology, and in particular to a method and system for local fine-tuning of shoe lasts based on deep learning. Background Technology

[0002] As the master mold for shoemaking, the shape of the shoe last directly determines the fit, comfort, and functionality of the footwear. With consumers' increasing demand for personalized and functional footwear, traditional shoe last design methods face numerous challenges.

[0003] At the data acquisition level, current technologies primarily rely on static 3D scanning or 2D size measurement of the foot. This method can only capture the geometric shape of the foot in a non-weight-bearing or static standing posture, failing to reflect the true biomechanical state of the foot's bones, muscles, and soft tissues during dynamic movements such as walking, running, and climbing. For example, it cannot reflect the dynamic undulations of the arch, the flexion and extension of the forefoot, or the real-time changes in pressure distribution. Therefore, shoe lasts designed based on static data struggle to meet users' precise functional needs for support, cushioning, and flexibility in actual sports scenarios.

[0004] In terms of design methodology, traditional shoe last design relies heavily on the designer's experience and subjective judgment. Optimization is achieved through a cycle of prototyping, user trials, and feedback-based modifications, a lengthy and costly process. While some computer-aided design systems have been introduced, their core remains the same: scaling up a basic last or making local adjustments based on limited rules. They lack the data-driven analysis and quantitative modeling capabilities to address the real needs of users across multiple dimensions and scenarios. For footwear that needs to accommodate various sports activities (such as commuting and weekend hikes), designers often struggle to scientifically balance functional conflicts in different scenarios, leading to design compromises and failing to achieve optimal performance for "multi-functional shoes."

[0005] In terms of modular design, some existing adjustable shoes or insoles often have modular sections that are pre-fixed and subjectively set, rather than based on user-specific biomechanical data and multi-scenario demand analysis. This "one-size-fits-all" modular strategy may lead to unnecessary structural complexity and weight added to areas that do not require adjustment, while failing to provide effective customized solutions for critical areas where functional conflicts exist, and the adjustment of modules lacks precise data guidance.

[0006] In summary, existing technologies have the following limitations: 1. The data foundation is static, failing to capture dynamic, multi-scenario biomechanical characteristics; 2. The design process relies on experience, lacking objective and quantitative demand analysis and conflict resolution mechanisms; 3. The modularization strategy is indiscriminate, failing to achieve data-driven, precisely targeted "on-demand modularization." Therefore, there is an urgent need for a shoe last design system that integrates dynamic data acquisition, intelligent demand analysis, conflict identification, and precise modular design to scientifically and efficiently achieve personalized and functional fine-tuning of shoe lasts, meeting the complex and diverse usage needs of modern consumers. Summary of the Invention

[0007] To address the shortcomings of existing technologies, this application provides a method and system for local fine-tuning of shoe lasts based on deep learning.

[0008] Firstly, this application provides a deep learning-based shoe last local fine-tuning system, comprising: The data acquisition module is used to acquire dynamic foot data of the user under various preset sports scenarios. The dynamic foot data includes at least a time-varying sequence of plantar pressure distribution data and a three-dimensional contour data sequence of the foot. A multi-scenario analysis module, connected to the data acquisition module, is used to receive the foot dynamic data, extract biomechanical feature vectors for each preset sports scenario based on the foot dynamic data, and analyze the biomechanical feature vectors using a clustering algorithm to generate functional requirement profiles corresponding to different sports modes. The functional requirement profiles are used to quantitatively describe the intensity of the demand for different functional dimensions of the shoe last in the corresponding sports mode. The functional conflict identification module is connected to the multi-scene analysis module. It is used to receive the functional requirement profile, calculate the functional requirement difference between different sports modes, and compare the functional requirement difference with a preset conflict threshold to identify local areas on the shoe last where there is a functional requirement conflict. The modular partitioning and output module is connected to the functional conflict identification module and is used to generate a modular partitioning scheme for shoe lasts and a corresponding adjustable module parameter set based on the identified local areas where there are functional conflict.

[0009] Preferably, the data acquisition module specifically includes: The pressure sensing unit is used to acquire the foot pressure distribution data sequence corresponding to the user under various preset motion scenarios. The foot pressure distribution data sequence is a set of pressure values ​​of each area of ​​the foot collected in a predetermined time sequence. The three-dimensional contour capture unit is used to acquire the three-dimensional contour data sequence of the user's foot under various preset motion scenarios. The three-dimensional contour data sequence of the foot is the three-dimensional point cloud data of the foot surface that is collected synchronously with the pressure sensing unit and changes according to a predetermined time sequence. When the multi-scene analysis module extracts the biomechanical feature vector, it associates the plantar pressure distribution data and the three-dimensional contour data of the foot at the same moment to calculate a composite feature including the trajectory of the pressure center, the joint movement angle, and the change in foot shape.

[0010] Preferably, in the multi-scenario analysis module, a clustering algorithm is used to analyze the biomechanical feature vectors, specifically including: The biomechanical feature vectors are input into a preset hierarchical clustering model, which clusters different preset motion scenarios into several motion pattern clusters based on feature similarity. For each motion mode cluster, the biomechanical feature vectors of all preset motion scenarios contained therein are aggregated, and the statistical values ​​of the demand intensity of each functional dimension are calculated to generate the functional demand profile corresponding to the motion mode cluster.

[0011] Preferably, the functional conflict identification module calculates the functional requirement difference degree by including: Construct a functional requirement matrix, where rows of the functional requirement matrix represent different sports mode clusters, columns represent preset shoe last functional dimensions, and matrix element values ​​are the requirement intensity of the functional dimension in the functional requirement profile of the corresponding sports mode cluster. Calculate the Euclidean distance between any two row vectors in the functional requirements matrix, and use the calculation result as the functional requirements difference degree between the corresponding two motion mode clusters.

[0012] Preferably, the functional conflict identification module identifies local areas where functional requirements conflict, specifically including: When the difference in functional requirements of two movement pattern clusters for the same shoe last functional dimension exceeds a first preset threshold, and the functional dimension is mapped to a specific physical area of ​​the shoe last, the specific physical area is marked as a potential conflict area. Based on the frequency data of each sports mode provided by the user, the severity of the conflict in the potential conflict area is weighted and calculated. Potential conflict areas whose weighted calculation results exceed a second preset threshold are identified as local areas with functional requirement conflicts.

[0013] Preferably, when the modular partitioning and output module generates the modular partitioning scheme for the shoe last, the principle is: modular design is only implemented in local areas where the functional conflict identification module determines that there is a functional requirement conflict; for areas where no functional requirement conflict is identified, an integrated shoe last structure design is adopted.

[0014] Preferred options also include: The combined recommendation module is connected to the modular partitioning and output module and the functional conflict identification module, and is used to receive target scenario usage frequency distribution data input by the user, as well as the adjustable module parameter set; The combination recommendation module calculates the weight of each motion mode cluster based on the frequency distribution data of the target scene, and selects the optimal combination of module parameters for each modular partition from the adjustable module parameter set based on the weight, generating and outputting a personalized module combination recommendation scheme for the user.

[0015] Preferred options also include: The verification and feedback module is used to receive user feedback data on their experience using the personalized module combination recommendation scheme. The verification and feedback module compares the experience feedback data with the predicted performance data, generates parameter optimization suggestions, and feeds the parameter optimization suggestions back to the modular partitioning and output module for iteratively updating the adjustable module parameter set.

[0016] Secondly, this application provides a deep learning-based method for local fine-tuning of shoe lasts, including: Acquire dynamic foot data of users under various simulated preset sports scenarios. The dynamic foot data includes at least a time-varying sequence of plantar pressure distribution data and a sequence of three-dimensional foot contour data. The foot dynamic data is received, and biomechanical feature vectors for each preset sports scenario are extracted based on the foot dynamic data. The biomechanical feature vectors are analyzed using a clustering algorithm to generate functional requirement profiles corresponding to different sports modes. The functional requirement profiles are used to quantitatively describe the intensity of the demand for different functional dimensions of the shoe last in the corresponding sports mode. Receive the functional requirement profile, calculate the functional requirement difference between different sports modes, and compare the functional requirement difference with a preset conflict threshold to identify local areas on the shoe last where there is a functional requirement conflict. Based on the identified local areas with conflicting functional requirements, a modular partitioning scheme for shoe lasts and a corresponding set of adjustable module parameters are generated.

[0017] Thirdly, this application provides a computer-readable storage medium storing instructions that, when executed on a computer, cause the computer to perform any of the above-described deep learning-based shoe last local fine-tuning systems.

[0018] In summary, this application includes at least one of the following beneficial technical effects: 1. This application provides a deep learning-based shoe last local fine-tuning system. Through a data acquisition module, it simulates various preset sports scenarios and simultaneously acquires the plantar pressure distribution sequence and the three-dimensional contour sequence of the foot. It systematically captures the biomechanical dynamic characteristics of the foot under real sports conditions, thereby changing the limitations of traditional reliance on static, non-load-bearing data. It provides a high-fidelity data foundation for shoe last design that reflects the actual functional state of the foot, enabling the design to closely meet the real needs of users in dynamic processes such as walking and running. 2. The multi-scenario analysis module uses clustering algorithms to intelligently analyze multi-dimensional biomechanical characteristics, automatically categorizes complex scenarios into different movement modes, and generates a quantitative "functional requirement profile". This reduces the over-reliance on the designer's subjective experience in traditional design. Through a data-driven approach, it objectively and quantitatively identifies and defines the specific demand intensity of different movement modes on each functional dimension of the shoe last, making design decisions based on evidence. 3. The functional conflict identification module calculates the difference in needs between different sports modes and combines it with the frequency of user use. It can automatically and accurately identify specific local areas on the shoe last where there are functional conflict needs. This provides a clear focus of conflict and a basis for trade-offs, so that when designing shoes that take into account multiple scenarios, blind compromises can be avoided and targeted scientific optimization can be carried out instead. 4. Modular partitioning and output modules: Based on the conflict identification results, modular design is implemented only in key areas where there is a genuine functional conflict. For areas without conflict, the integrated structure is retained. The data-driven and precise partitioning strategy overcomes the structural redundancy and weight increase problems caused by the existing "one-size-fits-all" modularization. It ensures that each part of the modular design directly corresponds to a clear and verified user need conflict, thereby maximizing the structural efficiency and overall performance of the product while realizing personalized functional adjustments. 5. Through the combination recommendation module, the system can automatically recommend the optimal combination of module parameters based on the user's individual usage habits, providing a highly personalized solution. Further verification and feedback modules collect actual usage data, compare it with the predictive model, and iteratively optimize the parameter set. This forms a complete closed loop of "data collection - intelligent analysis - precise design - personalized recommendation - feedback optimization," significantly shortening the traditional lengthy "try-on - feedback - modification" cycle, reducing customization costs, and ultimately improving product adaptability and user comfort through continuous iteration. Attached Figure Description

[0019] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a schematic diagram of a system for local fine-tuning of shoe lasts based on deep learning, according to an embodiment of this application.

[0021] Figure 2 This is a flowchart of a method for local fine-tuning of shoe lasts based on deep learning, according to an embodiment of this application. Detailed Implementation

[0022] The following is in conjunction with the appendix Figure 1-2 This application will be described in further detail.

[0023] Example 1 This application discloses a deep learning-based shoe last local fine-tuning system.

[0024] Reference Figure 1 A deep learning-based shoe last local fine-tuning system includes: The data acquisition module is used to acquire dynamic foot data of the user under various preset sports scenarios. The dynamic foot data includes at least a time-varying sequence of plantar pressure distribution data and a three-dimensional contour data sequence of the foot. In this embodiment, the data acquisition module is the basic data input link of the entire system. Its function is to capture the dynamic biomechanical characteristics of the foot during movement, rather than the traditional static dimensions. Specifically, by simulating real sports scenarios, the system can obtain key time-series data such as how plantar pressure is transferred and how the foot shape deforms. These data are the original basis for subsequent analysis of functional conflict requirements. Through the above operations, it is ensured that all subsequent analyses are based on real dynamic data that is strongly related to the scenario, thereby significantly improving the practicality and accuracy of the shoe last design scheme.

[0025] Furthermore, the data acquisition module simulates various preset motion scenarios, including but not limited to walking on flat ground, going uphill on a slope, going downhill on a slope, and rock climbing and gripping scenarios, and the acquisition of the foot dynamic data is carried out synchronously on a simulation platform with adjustable slope and surface material.

[0026] A multi-scenario analysis module, connected to the data acquisition module, is used to receive the foot dynamic data, extract biomechanical feature vectors for each preset sports scenario based on the foot dynamic data, and analyze the biomechanical feature vectors using a clustering algorithm to generate functional requirement profiles corresponding to different sports modes. The functional requirement profiles are used to quantitatively describe the intensity of the demand for different functional dimensions of the shoe last in the corresponding sports mode. Specifically, the multi-scenario analysis module is the core computing engine of the system, responsible for dimensionality reduction, abstraction, and pattern recognition of the raw data. It transforms time-series data into feature vectors representing the essential characteristics of each scenario, and through clustering, discovers "movement patterns" with similar biomechanical requirements hidden behind scenarios with different names. The technical effect is to achieve a mapping from specific "scenarios" to abstract "functional requirements," providing structured input for subsequent functional conflict analysis. For example, it might find that "climbing an incline" and "climbing stairs" are clustered into the same pattern, as they have highly similar requirements for forefoot propulsion and ankle locking.

[0027] The functional conflict identification module is connected to the multi-scene analysis module. It is used to receive the functional requirement profile, calculate the functional requirement difference between different sports modes, and compare the functional requirement difference with a preset conflict threshold to identify local areas on the shoe last where there is a functional requirement conflict. Specifically, the functional conflict identification module is the system's decision-making and positioning center. Its role is to automatically identify irreconcilable design contradictions between different movement modes based on quantified functional requirements, and precisely locate these contradictions in specific physical areas of the shoe last. Its technical effect is to transform the traditional experience-based functional trade-off process into an objective, data-driven automated process, thereby accurately identifying key areas requiring physical structural adjustments. For example, calculations might reveal that "long-distance hiking mode" requires high arch support to alleviate fatigue, while "track running mode" requires the arch area to have certain energy rebound characteristics. These two requirements directly conflict in the rigidity and structural design of the arch support module.

[0028] The modular partitioning and output module is connected to the functional conflict identification module and is used to generate a modular partitioning scheme for shoe lasts and a corresponding adjustable module parameter set based on the identified local areas where there are functional conflict.

[0029] Specifically, the modular partitioning and output module is the system's design output end. Its function is to transform the analytical conclusion of "conflict identification" into a concrete physical design scheme that can guide production. It defines which areas on the shoe last should be replaceable / adjustable modules and their interface standards, and provides detailed physical parameters for each module. This enables a direct and automated output from "problem diagnosis" (conflict identification) to "solution" (modular design), making the design scheme highly executable. For example, based on the aforementioned identified arch area conflict, this module will output a design drawing, defining the arch area as an independent pluggable support module area, and providing detailed stiffness, curvature, and thickness parameters for two support modules adapted for "hiking" and "running," respectively.

[0030] For example, the data acquisition module specifically includes: The pressure sensing unit is used to acquire the foot pressure distribution data sequence corresponding to the user under various preset motion scenarios. The foot pressure distribution data sequence is a set of pressure values ​​of each area of ​​the foot collected in a predetermined time sequence. Specifically, the pressure sensing unit digitizes the mechanical interaction between the sole of the foot and the support surface, forming a quantifiable sequence of pressure maps. This provides a direct data source for assessing local load-bearing capacity, impact, and stability. For example, pressure data can clearly show that the forefoot metatarsal region experiences abnormally concentrated pressure in a rock climbing grip posture, providing crucial input for subsequently identifying the functional requirements of this area.

[0031] The three-dimensional contour capture unit is used to acquire the three-dimensional contour data sequence of the user's foot under various preset motion scenarios. The three-dimensional contour data sequence of the foot is the three-dimensional point cloud data of the foot surface that is collected synchronously with the pressure sensing unit and changes according to a predetermined time sequence. Specifically, the 3D contour capture unit accurately records the spatial morphological changes of the foot's soft tissues and skeletal joints during movement. Its technical advantage lies in compensating for the shortcomings of pressure data in spatial morphological representation, enabling the system to analyze the actual space occupied by the foot within the shoe cavity and its movement trajectory. Through strict time synchronization between the 3D contour capture unit and the pressure sensing unit, it is ensured that each frame of pressure data is accurately correlated with the corresponding 3D morphology of the foot, thereby allowing the calculation of "pressure-morphology" composite features, such as the degree of arch collapse under specific pressure.

[0032] When the multi-scene analysis module extracts the biomechanical feature vector, it associates the plantar pressure distribution data and the three-dimensional contour data of the foot at the same moment to calculate a composite feature including the trajectory of the pressure center, the joint movement angle, and the change in foot shape.

[0033] In this embodiment, the system can calculate an "ankle stability coefficient" by correlating the center of pressure and the ankle joint angle; and can assess "propulsion efficiency" by correlating the peak forefoot pressure and the metatarsophalangeal joint flexion angle. These composite characteristics are key indicators that have never been considered in traditional shoe last design.

[0034] For example, in the multi-scenario analysis module, a clustering algorithm is used to analyze the biomechanical feature vectors, specifically including: The biomechanical feature vectors are input into a preset hierarchical clustering model, which clusters different preset motion scenarios into several motion pattern clusters based on feature similarity. Specifically, hierarchical clustering models construct a tree-like hierarchy of scenes from bottom to top or top to bottom based on the distance between feature vectors, allowing the system to segment different motion pattern clusters according to a preset distance threshold. The advantage of using hierarchical clustering lies in its intuitive process and its ability to demonstrate the relationships between different scenes. As an alternative technique, K-means clustering or density clustering algorithms can also be used, the choice depending on the characteristics of the data distribution and whether the number of clusters needs to be pre-specified.

[0035] For each motion mode cluster, the biomechanical feature vectors of all preset motion scenarios contained therein are aggregated, and the statistical values ​​of the demand intensity of each functional dimension are calculated to generate the functional demand profile corresponding to the motion mode cluster.

[0036] By adopting the above technical solution, noise that may exist in individual scene data is eliminated, making the demand profile more representative and robust. For example, for a pattern cluster containing "rock climbing" and "tree climbing", the system will calculate the average demand intensity of all data in the cluster in dimensions such as "forefoot flexion freedom" and "metatarsal support stiffness", thereby generating a demand profile representing "high flexibility climbing mode".

[0037] For example, the functional conflict identification module calculates the functional requirement difference degree by including: Construct a functional requirement matrix, where rows of the functional requirement matrix represent different sports mode clusters, columns represent preset shoe last functional dimensions, and matrix element values ​​are the requirement intensity of the functional dimension in the functional requirement profile of the corresponding sports mode cluster. Specifically, this step involves structuring the data before conflict quantification. Its purpose is to transform the abstract demand profile into a standard mathematical matrix representation, making the demand intensity of different movement modes across various functional dimensions immediately apparent, facilitating subsequent numerical calculations and comparisons. This matrix is ​​the key data structure connecting "demand analysis" and "conflict identification." For example, a simplified matrix might look like this: Rows: Mode A (hiking), Mode B (running); Columns: Arch support, forefoot flexibility; Matrix element values ​​are ratings from 1 to 10.

[0038] Calculate the Euclidean distance between any two row vectors in the functional requirements matrix, and use the calculation result as the functional requirements difference degree between the corresponding two motion mode clusters.

[0039] Specifically, the Euclidean distance is used to calculate a scalar difference value for any two movement modes, which comprehensively reflects their overall difference in requirements across all functional dimensions. The calculation formula is: in and Let D represent the demand intensity of Mode A and Mode B in the i-th functional dimension, respectively, where n is the total number of functional dimensions. Taking a simplified matrix as an example, if the demand vector for Mode A is [8,2] (support 8, flexibility 2) and for Mode B it is [3,7], then the functional demand difference D... E ≈7.07, this value will be used for subsequent conflict threshold judgment.

[0040] For example, the functional conflict identification module identifies local areas where functional requirements conflict, specifically including: When the difference in functional requirements of two movement pattern clusters for the same shoe last functional dimension exceeds a first preset threshold, and the functional dimension is mapped to a specific physical area of ​​the shoe last, the specific physical area is marked as a potential conflict area. Specifically, the first preset threshold serves to establish a sensitivity threshold, filtering out minor, negligible, or easily mitigated differences in requirements through conventional means such as material elasticity, focusing only on sharp, fundamental design conflicts. For example, setting the first preset threshold to 4 means that the system only considers a potential conflict when the difference in requirement scores between the two modes for "forefoot flexibility" exceeds 4. Subsequently, the system uses a built-in "function-physical area" mapping relationship (e.g., mapping "forefoot flexibility" to "thickness and curvature of the metatarsophalangeal joint area of ​​the shoe last") to pinpoint the conflict to a specific physical area of ​​the shoe last.

[0041] Based on the frequency data of each sports mode provided by the user, the severity of the conflict in the potential conflict area is weighted and calculated. Potential conflict areas whose weighted calculation results exceed a second preset threshold are identified as local areas with functional requirement conflicts.

[0042] Specifically, this step introduces user personalization factors, refining and customizing conflict identification. Its purpose is to assess the actual impact of "potential conflicts" on specific users, ensuring that the final modular design prioritizes resolving core conflicts encountered in the most frequent user scenarios. The technical effect is to make the design more targeted. The calculation formula can be: Conflict Severity = Functional Requirement Difference × (Mode A Usage Frequency × Mode B Usage Frequency). Using the previous example, if the user's input frequency is 60% walking and 40% running, then the weighted severity ≈ 7.07 × (0.6 × 0.4) ≈ 1.70. The level of the second preset threshold (e.g., set to 1.0) directly controls the "aggressiveness" of the system's modular design: the lower the threshold, the more areas are judged to require modularization; the higher the threshold, the more the system tends towards an integrated design.

[0043] For example, when the modular partitioning and output module generates the modular partitioning scheme for the shoe last, the principle is: modular design is only implemented in local areas where the functional conflict identification module determines that there is a functional requirement conflict; for areas where no functional requirement conflict is identified, an integrated shoe last structure design is adopted.

[0044] Specifically, the aforementioned implementation process fundamentally optimizes the logic of modular design, avoiding modularization for its own sake or the use of pre-fixed, subjective partitioning schemes. Its technical effect lies in its ability to provide adaptability to multiple scenarios while maximizing the overall structural integrity, lightweight design, and reliability of the shoe last, thus achieving the optimal balance between functional customization and basic product performance. Connecting to the previous steps, if the demand intensity of the "heel stability area" is highly consistent (low variation) across all sports modes, with no significant conflicts, then this area will be designed as a robust, integrated structure, rather than an unnecessary replaceable module.

[0045] For example, it also includes: The combined recommendation module is connected to the modular partitioning and output module and the functional conflict identification module, and is used to receive target scenario usage frequency distribution data input by the user, as well as the adjustable module parameter set; Specifically, the combination recommendation module provides a personalized service interface for end users. Its function is to allow users to easily input their actual usage habits (frequency distribution of scenarios), and based on this, provide tailored module configuration suggestions, greatly reducing the difficulty of selection and combination for users and improving the system's usability and practicality. For example, the user interface provides a slider for users to set "70% city commuting" and "30% weekend hiking".

[0046] The combination recommendation module calculates the weight of each motion mode cluster based on the frequency distribution data of the target scene, and selects the optimal combination of module parameters for each modular partition from the adjustable module parameter set based on the weight, generating and outputting a personalized module combination recommendation scheme for the user.

[0047] Specifically, this step is the core of the computation for personalized recommendations. Its function is to use optimization algorithms to find the "optimal solution" for module configurations that best meet the user's needs across various scenarios. One feasible implementation is a weighted scoring method: for each modular partition, calculate the weighted comprehensive score of each candidate module across all motion modes. ,in It is the usage frequency weight of the i-th motion pattern. The degree to which the parameters of the j-th candidate module meet the requirements of the i-th motion mode (derived from the matching degree calculation between the functional requirement profile and the module parameters), the system selects a comprehensive score S for each partition. j The highest-ranking candidate modules are combined to form the final recommended solution.

[0048] For example, it also includes: The verification and feedback module is used to receive user feedback data on their experience using the personalized module combination recommendation scheme. The verification and feedback module compares the experience feedback data with the predicted performance data, generates parameter optimization suggestions, and feeds the parameter optimization suggestions back to the modular partitioning and output module for iteratively updating the adjustable module parameter set.

[0049] Specifically, by employing the above technical solution, the system's core knowledge base—the adjustable module parameter set—is continuously calibrated and optimized using real-world feedback, thereby enhancing the system's recommendation capabilities as usage data accumulates. For example, if a large number of users report that "the arch support recommended for urban jogging feels too soft," the system will analyze the common data characteristics of these users and may generate suggestions such as increasing the standard hardness parameter of the arch module corresponding to the "urban jogging mode cluster," or adding a "high support variant" option to the parameter set.

[0050] Example 2 This application also discloses a method for local fine-tuning of shoe lasts based on deep learning.

[0051] Reference Figure 2 A deep learning-based method for local fine-tuning of shoe lasts includes: Acquire dynamic foot data of users under various simulated preset sports scenarios. The dynamic foot data includes at least a time-varying sequence of plantar pressure distribution data and a sequence of three-dimensional foot contour data. The foot dynamic data is received, and biomechanical feature vectors for each preset sports scenario are extracted based on the foot dynamic data. The biomechanical feature vectors are analyzed using a clustering algorithm to generate functional requirement profiles corresponding to different sports modes. The functional requirement profiles are used to quantitatively describe the intensity of the demand for different functional dimensions of the shoe last in the corresponding sports mode. Receive the functional requirement profile, calculate the functional requirement difference between different sports modes, and compare the functional requirement difference with a preset conflict threshold to identify local areas on the shoe last where there is a functional requirement conflict. Based on the identified local areas with conflicting functional requirements, a modular partitioning scheme for shoe lasts and a corresponding set of adjustable module parameters are generated.

[0052] The above content is merely an example and illustration of the concept of the present invention. Those skilled in the art can make various modifications or additions to the specific embodiments described or use similar methods to replace them, as long as they do not deviate from the concept of the invention, they should all fall within the protection scope of the present invention.

[0053] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0054] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to any specific implementation. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention.

Claims

1. A deep learning-based system for local fine-tuning of shoe lasts, characterized in that, include: The data acquisition module is used to acquire dynamic foot data of the user under various preset sports scenarios. The dynamic foot data includes at least a time-varying sequence of plantar pressure distribution data and a three-dimensional contour data sequence of the foot. A multi-scenario analysis module, connected to the data acquisition module, is used to receive the foot dynamic data, extract biomechanical feature vectors for each preset sports scenario based on the foot dynamic data, and analyze the biomechanical feature vectors using a clustering algorithm to generate functional requirement profiles corresponding to different sports modes. The functional requirement profiles are used to quantitatively describe the intensity of the demand for different functional dimensions of the shoe last in the corresponding sports mode. The functional conflict identification module is connected to the multi-scene analysis module. It is used to receive the functional requirement profile, calculate the functional requirement difference between different sports modes, and compare the functional requirement difference with a preset conflict threshold to identify local areas on the shoe last where there is a functional requirement conflict. The modular partitioning and output module is connected to the functional conflict identification module and is used to generate a modular partitioning scheme for shoe lasts and a corresponding adjustable module parameter set based on the identified local areas where there are functional conflict.

2. The deep learning-based shoe last local fine-tuning system according to claim 1, characterized in that, The data acquisition module specifically includes: The pressure sensing unit is used to acquire the foot pressure distribution data sequence corresponding to the user under various preset motion scenarios. The foot pressure distribution data sequence is a set of pressure values ​​of each area of ​​the foot collected in a predetermined time sequence. The three-dimensional contour capture unit is used to acquire the three-dimensional contour data sequence of the user's foot under various preset motion scenarios. The three-dimensional contour data sequence of the foot is the three-dimensional point cloud data of the foot surface that is collected synchronously with the pressure sensing unit and changes according to a predetermined time sequence. When the multi-scene analysis module extracts the biomechanical feature vector, it associates the plantar pressure distribution data and the three-dimensional contour data of the foot at the same moment to calculate a composite feature including the trajectory of the pressure center, the joint movement angle, and the change in foot shape.

3. The deep learning-based shoe last local fine-tuning system according to claim 2, characterized in that, In the multi-scenario analysis module, a clustering algorithm is used to analyze the biomechanical feature vectors, specifically including: The biomechanical feature vectors are input into a preset hierarchical clustering model, which clusters different preset motion scenarios into several motion pattern clusters based on feature similarity. For each motion mode cluster, the biomechanical feature vectors of all preset motion scenarios contained therein are aggregated, and the statistical values ​​of the demand intensity of each functional dimension are calculated to generate the functional demand profile corresponding to the motion mode cluster.

4. The deep learning-based shoe last local fine-tuning system according to claim 1, characterized in that, The functional conflict identification module calculates the functional requirement difference degree as follows: Construct a functional requirement matrix, where rows of the functional requirement matrix represent different sports mode clusters, columns represent preset shoe last functional dimensions, and matrix element values ​​are the requirement intensity of the functional dimension in the functional requirement profile of the corresponding sports mode cluster. Calculate the Euclidean distance between any two row vectors in the functional requirements matrix, and use the calculation result as the functional requirements difference degree between the corresponding two motion mode clusters.

5. The deep learning-based shoe last local fine-tuning system according to claim 1, characterized in that, The functional conflict identification module identifies local areas where functional requirements conflict, specifically including: When the difference in functional requirements of two movement pattern clusters for the same shoe last functional dimension exceeds a first preset threshold, and the functional dimension is mapped to a specific physical area of ​​the shoe last, the specific physical area is marked as a potential conflict area. Based on the frequency data of each sports mode provided by the user, the severity of the conflict in the potential conflict area is weighted and calculated. Potential conflict areas whose weighted calculation results exceed a second preset threshold are identified as local areas with functional requirement conflicts.

6. The deep learning-based shoe last local fine-tuning system according to claim 1, characterized in that, When the modular partitioning and output module generates the modular partitioning scheme for the shoe last, the principle is as follows: modular design is only implemented in local areas where the functional conflict identification module determines that there is a functional requirement conflict. For areas where no functional requirement conflict is identified, an integrated shoe last structure design is adopted.

7. The deep learning-based shoe last local fine-tuning system according to claim 1, characterized in that, Also includes: The combined recommendation module is connected to the modular partitioning and output module and the functional conflict identification module, and is used to receive target scenario usage frequency distribution data input by the user, as well as the adjustable module parameter set; The combination recommendation module calculates the weight of each motion mode cluster based on the frequency distribution data of the target scene, and selects the optimal combination of module parameters for each modular partition from the adjustable module parameter set based on the weight, generating and outputting a personalized module combination recommendation scheme for the user.

8. A deep learning-based shoe last local fine-tuning system according to claim 7, characterized in that, Also includes: The verification and feedback module is used to receive user feedback data on their experience using the personalized module combination recommendation scheme. The verification and feedback module compares the experience feedback data with the predicted performance data, generates parameter optimization suggestions, and feeds the parameter optimization suggestions back to the modular partitioning and output module for iteratively updating the adjustable module parameter set.

9. A deep learning-based method for local fine-tuning of shoe lasts, applied to the deep learning-based shoe last local fine-tuning system described in any one of claims 1-8, characterized in that, include: Acquire user foot dynamic data under simulated multiple preset sports scenarios, wherein the foot dynamic data includes at least a time-varying sequence of plantar pressure distribution data and a sequence of three-dimensional foot contour data; The foot dynamic data is received, and biomechanical feature vectors for each preset sports scenario are extracted based on the foot dynamic data. A clustering algorithm is used to analyze the biomechanical feature vectors to generate functional requirement profiles corresponding to different sports modes. The functional requirement profiles are used to quantitatively describe the intensity of the demand for different functional dimensions of the shoe last in the corresponding sports mode. Receive the functional requirement profile, calculate the functional requirement difference between different sports modes, and compare the functional requirement difference with a preset conflict threshold to identify local areas on the shoe last where there is a functional requirement conflict. Based on the identified local areas with conflicting functional requirements, a modular partitioning scheme for shoe lasts and a corresponding set of adjustable module parameters are generated.

10. A computer-readable storage medium, characterized in that: The system stores instructions that, when executed on a computer, cause the computer to perform a deep learning-based shoe last local fine-tuning system as described in any one of claims 1 to 8.

Images