Intelligent ornamentation pattern classification and recommendation system

Abstract

The invention discloses an intelligent ornamentation pattern classification and recommendation system, and relates to the field of computer systems. The intelligent decoration pattern classification and recommendation system is composed of a user terminal, a feature extraction server, a central processing server and a pattern database. The user terminal serves as an entrance for direct interaction with a user and is responsible for obtaining original data including a human body image, and the original data is a basic data source of the whole system. According to the intelligent ornamentation pattern classification and recommendation system, on the recommendation precision, the high precision of personalized recommendation is ensured through comprehensive consideration of multi-dimensional factors and application of a scientific algorithm; in the aspect of privacy protection, the innovative application of the federated learning framework provides a solid guarantee for user data security; in the aspect of fashion trend response, the dynamic weight adjustment module enables the system to closely follow the trend change and adjust the recommendation strategy in time.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computer systems, in particular to an intelligent pattern classification and recommendation system. BACKGROUND

[0002] At present, the application scenarios of patterns are increasingly diversified, from clothing prints in daily dressing to pattern design in the field of beauty, to pattern customization of personalized accessories. Users' demand for "pattern adaptation to themselves" is becoming stronger, not only requiring patterns to conform to personal aesthetic style, but also matching body shape characteristics, use scenarios and current fashion trends.

[0003] In the existing pattern recommendation technology, the following problems exist: First, most systems only recommend patterns based on users' subjective style preferences, without considering the adaptation relationship between human body shape characteristics (such as shoulder width, waist-hip ratio) and pattern layout. For example, recommending horizontal expansion patterns for users with high shoulder width ratio can magnify the body's shortcomings, and recommending dense patterns for slim users may result in visual clutter, which cannot optimize the visual presentation of body shape through patterns.

[0004] Second, the popularity evaluation of existing systems mostly relies on fixed databases or periodically updated static data. When fashion trends change rapidly, the system cannot adjust the recommendation weight in time, resulting in a disconnection between the recommended results and the current trend, and reducing user acceptance.

[0005] Third, some systems need to collect complete human body images of users and upload them to centralized servers for feature extraction, which poses a risk of original data leakage. Systems that focus on privacy protection lack sufficient user feature data, making it difficult to achieve personalized recommendation iteration. In view of the deficiencies of the prior art, the present application provides an intelligent pattern classification and recommendation system to solve the above problems. SUMMARY

[0006] In view of the deficiencies of the prior art, the present application provides an intelligent pattern classification and recommendation system. In terms of recommendation accuracy, the comprehensive consideration of multi-dimensional factors and the use of scientific algorithms ensure the high accuracy of personalized recommendation. In terms of privacy protection, the innovative application of the federated learning framework provides a solid guarantee for user data security. In terms of fashion trend response, the dynamic weight adjustment module enables the system to follow the trend changes and adjust the recommendation strategy in time.

[0007] To achieve the above purposes, the present application realizes the following technical solutions: an intelligent pattern classification and recommendation system, comprising: a user terminal for obtaining original data containing at least a human body image input by a user; a feature extraction server in communication connection with the user terminal, comprising: a posture feature extraction module configured to extract human posture key points of a user from the human body image and calculate body shape parameters including shoulder width and waist-hip ratio based on the key points; a style analysis module configured to analyze a dressing style label specified by the user or identified from historical images of the user; a central processing server in communication with the feature extraction server and the pattern database, comprising: a pattern matching degree calculation module configured to calculate a matching degree of each pattern in the pattern database with the user according to the body shape parameters and the dressing style label, wherein a layout sensitivity factor λ (0.6≤λ≤1.2) is introduced in the calculation process, which is dynamically adjusted based on the body shape parameters and used to control the weight of the layout suggestion on the final matching degree; a recommendation list generation module configured to generate a personalized pattern recommendation list according to the matching degree sorting; the pattern database configured to store pattern data and its associated attribute labels, wherein the attribute labels at least include pattern style, applicable part, layout orientation and popularity index.

[0008] Preferably, the pattern matching degree calculation module calculates the matching degree Score by executing the following algorithm: Score=α×S_s+β×S_t+λ×S_l wherein S_s is a style score based on the similarity between the user dressing style label and the pattern style label, S_t is a popularity score based on real-time trend data, S_l is a layout score based on the compliance between the body shape parameters and the pattern layout orientation, α and β are configurable weight coefficients, and α+β+λ=1, wherein λ is the layout sensitivity factor.

[0009] Preferably, the layout sensitivity factor λ is dynamically set according to the ratio R_s of the user's shoulder width to height: when R_s is greater than a threshold T_high, λ is set to a maximum value λ_max (1.0≤λ_max≤1.2) to enhance the avoidance effect of unsuitable horizontal expansion patterns for users with high shoulder width; when R_s is less than a threshold T_low, λ is set to a minimum value λ_min (0.6≤λ_min≤0.8) to reduce the weight of layout suggestion.

[0010] Preferably, the central processing server further comprises a dynamic weight adjustment module configured to periodically crawl social media data to analyze current popular elements and dynamically adjust the value of the weight coefficient β accordingly. The faster the popular trend changes, the greater the adjustment amplitude Δβ of β.

[0011] Preferably, the posture feature extraction module adopts a neural network model trained based on a federated learning framework, which is locally trained on the user terminal or edge computing node, and only the update amount of model parameters is encrypted and uploaded to the feature extraction server for aggregation, and the original human body image data remains local.

[0012] Preferably, the local training period E_l of the federated learning framework is inversely related to the number of users N, and is set as E_l = max(1, round(C / N)), where C is a constant and round is an integer function, to ensure that the system can efficiently protect privacy and quickly converge when the scale of users expands.

[0013] Preferably, the system further comprises a collision prevention processing module for comparing the recommendation list with a public pool recording high-frequency use patterns, and introducing a penalty factor for a pattern in the recommendation list that has a matching degree higher than a threshold but also has a use frequency higher than a frequency threshold in the public pool, to reduce its final ranking.

[0014] Preferably, the penalty factor γ of the collision prevention processing module is calculated as: γ = 1 / (1+k×F), where F is the use frequency of the pattern in the public pool, and k is an inhibition coefficient greater than 0, and the avoidance strength of the system to the "collision" phenomenon is controlled by adjusting k (0.1≤k≤0.5).

[0015] Preferably, the user terminal is also used to receive the feedback operation of the user on the recommendation result; and the central processing server further comprises a feedback learning module for fine-tuning the weight coefficients α, β corresponding to the user according to the positive or negative feedback of the user, to realize iteration of the user's personalized recommendation strategy.

[0016] Preferably, the system further comprises a virtual try-on interface generation module connected with the recommendation list generation module, for superimposing and rendering the recommended pattern ranked first on the human body image uploaded by the user through augmented reality technology, to generate and output a virtual try-on effect picture.

[0017] It has the following beneficial effects: 1. The intelligent pattern classification and recommendation system obtains body shape parameters such as shoulder width and waist-hip ratio of the user through the posture feature extraction module, combines the dressing preference identified by the style analysis module, introduces a layout sensitivity factor λ that is dynamically adjusted, for example, automatically reduces the matching weight of horizontally expanded patterns for users with high shoulder width ratio, and preferentially recommends patterns of the same style and in line with current popular trends for "vintage" style users, to ensure that the recommendation result not only fits the user's body shape characteristics, but also meets the aesthetic and trend needs.

[0018] 2、The intelligent ornament pattern classification and recommendation system, the dynamic weight adjustment module equipped with the system can periodically capture social media data, deeply analyze current popular elements, and dynamically adjust the weight coefficient according to the analysis result. This mechanism enables the recommendation result to quickly respond to the rapid changes of fashion trends, ensuring that users can always access pattern selection that meets the current popular trends, improving user satisfaction and recognition of the recommendation result.

[0019] 3、The intelligent ornament pattern classification and recommendation system uses an advanced federated learning framework for pose feature extraction, keeping the original human image data locally and uploading only encrypted model parameter updates. This innovative approach effectively addresses user privacy concerns during data sharing, ensuring the normal operation and optimization of system functions while protecting user privacy. BRIEF DESCRIPTION OF DRAWINGS

[0020] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiment or prior art description. Obviously, the drawings in the following description are only some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor on the basis of these drawings.

[0021] Figure 1 is the overall architecture diagram of the system of the present application; Figure 2 is a detailed module diagram of the feature extraction server of the present application; Figure 3 is a detailed module diagram of the central processing server of the present application; Figure 4 is a pattern matching degree calculation flowchart of the present application; Figure 5 is a layout sensitivity factor λ dynamic setting logic diagram of the present application; Figure 6 is a collision avoidance processing module logic diagram of the present application. DETAILED DESCRIPTION

[0022] In order to make the purpose, technical scheme and advantages of the embodiments of the present application more clear, the technical scheme in the embodiments of the present application is described clearly and completely. Obviously, the described embodiments are part of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0023] In order to better understand the above technical solutions, the above technical solutions will be described in detail in combination with the drawings in the specification and specific embodiments.

[0024] The embodiment of the application discloses an intelligent ornament pattern classification and recommendation system. Figures 1-6 As shown in the figure, comprising: A user terminal is used to acquire original data containing a human body image input by a user; A feature extraction server is in communication connection with the user terminal, comprising: A posture feature extraction module is used to extract human body posture key points of the user from the human body image, and calculate body shape parameters including shoulder width and waist-hip ratio based on the key points; A style analysis module is used to analyze a dressing style label specified by the user or identified from historical images of the user; A central processing server is in communication connection with the feature extraction server and a pattern database respectively, comprising: A pattern matching degree calculation module is used to calculate matching degrees of patterns in the pattern database and the user according to the body shape parameters and the dressing style label; a layout sensitivity factor λ (0.6≤λ≤1.2) is introduced in the calculation process, which is dynamically adjusted based on the body shape parameters and used to control the weight of the influence of pattern layout suggestions on the final matching degree; A recommendation list generation module is used to generate a personalized pattern recommendation list according to the matching degrees; The pattern database is used to store pattern data and its associated attribute labels, and the attribute labels at least include pattern style, applicable part, layout orientation and popularity index.

[0025] The intelligent ornament pattern classification and recommendation system is composed of a user terminal, a feature extraction server, a central processing server and a pattern database. The user terminal serves as an entrance for direct interaction with the user and is responsible for acquiring original data containing a human body image, which is the basic data source of the whole system. The feature extraction server undertakes the task of extracting key features from the human body image. The posture feature extraction module thereof accurately locates human body posture key points from the human body image through a specific algorithm, and then calculates body shape parameters such as shoulder width and waist-hip ratio. These parameters can objectively reflect the physical characteristics of the user and provide an important basis for subsequent pattern matching. The style analysis module understands the aesthetic preference of the user by analyzing the dressing style label specified by the user or identified from historical images.

[0026] The central processing server is the core processing unit of the system, which communicates with the feature extraction server and the pattern database. The pattern matching degree calculation module calculates the matching degree of the pattern and the user according to the body shape parameters and the dressing style label, combined with the layout sensitivity factor. The introduction of the layout sensitivity factor enables the system to dynamically adjust the influence weight of the layout suggestion of the pattern on the matching degree according to the user's body shape, improving the accuracy of the recommendation. The recommendation list generation module generates a personalized pattern recommendation list according to the matching degree sorting, providing the user with a pattern selection that meets his / her characteristics and preferences. The pattern database stores rich pattern data and its associated attribute labels, such as pattern style, applicable part, layout orientation, and popularity index, etc. These attribute labels provide multi-dimensional reference information for pattern matching.

[0027] The system obtains data through the user terminal, the feature extraction server performs feature analysis, the central processing server combines the pattern database information to perform matching calculation and recommendation generation, forming a complete data processing and recommendation process.

[0028] The system architecture is clear, and each module has a clear division of labor, which can efficiently complete the whole process from data acquisition to recommendation generation, providing personalized pattern recommendation services for users.

[0029] The pattern matching degree calculation module calculates the matching degree Score by executing the following algorithm: Score=α×S_s+β×S_t+λ×S_l Wherein, S_s is the style score based on the similarity of user dressing style label and pattern style label, S_t is the popularity score based on real-time trend data, S_l is the layout score based on the conformity of body shape parameters and pattern layout orientation; α, β are configurable weight coefficients, and α+β+λ=1, wherein λ is the layout sensitivity factor.

[0030] The algorithm calculates the matching degree by considering the style score, the popularity score and the layout score. The style score is based on the similarity of user dressing style label and pattern style label, which can accurately reflect the degree of fit between the pattern and the user's aesthetic preference; the popularity score combines real-time trend data, making the recommendation results conform to the current fashion trend; the layout score ensures the display effect of the pattern on the user's body according to the conformity of user body shape parameters and pattern layout orientation. The configurable weight coefficients α, β and the layout sensitivity factor λ can be adjusted according to actual needs, making the matching degree calculation more flexible and accurate.

[0031] The system weights and sums the scores of different dimensions according to certain weights to get the comprehensive matching degree score. By considering multiple factors that affect pattern matching, the system can meet the needs of different users and scenarios through the adjustment of weight coefficients, improving the accuracy and personalization of the recommendation.

[0032] The layout sensitivity factor λ is dynamically set according to the ratio R_s of the user's shoulder width to height: when R_s is greater than a threshold T_high, λ is set to a maximum value λ_max (1.0 ≤ λ_max ≤ 1.2) to enhance the avoidance effect of the horizontal expansion pattern on users with high shoulder width; when R_s is less than a threshold T_low, λ is set to a minimum value λ_min (0.6 ≤ λ_min ≤ 0.8) to reduce the weight of the layout recommendation.

[0033] The layout sensitivity factor λ is dynamically set according to the ratio R_s of the user's shoulder width to height. When R_s is greater than a threshold T_high, it indicates that the user's shoulder width is relatively wide compared to height, at which time λ is set to a maximum value λ_max to enhance the avoidance effect of the horizontal expansion pattern on users with high shoulder width, avoiding recommending patterns that are not suitable for the user's body type; when R_s is less than a threshold T_low, it indicates that the user's shoulder width is relatively narrow compared to height, at which time λ is set to a minimum value λ_min to reduce the weight of the layout recommendation, allowing other factors to play a greater role in the matching degree calculation.

[0034] By setting different thresholds and corresponding λ values, the layout sensitivity factor is dynamically adjusted according to the ratio of the user's shoulder width to height, thereby affecting the calculation of the pattern matching degree. The system can make personalized layout recommendation adjustments according to the user's physical characteristics, improve the adaptability of the pattern to the user's body type, and enhance the practical application effect of the recommended results.

[0035] The central processing server further comprises a dynamic weight adjustment module for periodically crawling social media data to analyze current popular elements and dynamically adjust the value of the weight coefficient β accordingly. The faster the trend changes, the greater the adjustment range Δβ of β.

[0036] The core task of the dynamic weight adjustment module is to convert the abstract "trend" of social media into a quantifiable numerical parameter β value for controlling the recommendation algorithm, and to ensure that the system's response speed to the trend is positively related to the speed of the trend itself. The computer implementation includes the following steps: Step 1: Data collection and preprocessing The dynamic weight adjustment module has a built-in or called web crawler submodule, which automatically starts according to a preset period (e.g., every 6 hours).

[0037] The crawler will perform data crawling on a pre-defined set of target data sources, including: Targeted crawling of fashion blogger gathering places, trend information websites, and e-commerce platform trend communities (such as Xiaohongshu, Douyin, Weibo fashion super topic, Demao community, etc.).

[0038] Content scraping: Periodically scrape (e.g., every 4 hours) posts, videos, and image-text content containing specific keywords (e.g., "national trend," "new Chinese style," "street fashion," "print," "buckle," "Suzhou embroidery," etc.) with metadata, including: Text: Title, body, comments.

[0039] Tags: User-added #topic hashtags.

[0040] Interaction data: Likes, shares, comments, views.

[0041] Timestamp: Content publication time.

[0042] Step 2: Preprocess the data source: Clean and standardize the raw data (possibly in JSON or HTML format) scraped. This includes: de-duplication, removal of invalid characters, uniform data format (e.g., convert all times to UTC timestamps, normalize all numerical indicators to the [0,1] interval), and store in a time series database. Use NLP (Natural Language Processing) tool libraries and rule engines.

[0043] Step 3: Trend analysis and quantification: Use time series analysis methods. Aggregate relevant data for a certain pattern element (e.g., "cloud pattern") within a period (e.g., 24 hours) and calculate its heat value (H_t). The heat value can be a comprehensive indicator, for example: H_t=log((likes+2×shares+1.5×comments)×views^0.5)×keyword frequency.

[0044] Store the heat value of each element in each period to form a heat time series [H_t, H_t-1, H_t-2,...].

[0045] Calculate the trend change rate (a_t), which is the key to achieving "the faster the change, the greater the adjustment": a_t=(H_t-H_t-1) / ΔT, ΔT is the period interval time Step 4: Dynamically adjust the weight β: Establish a mapping relationship: Map the trend change rate a_t calculated in the previous step to the weight adjustment amplitude Δβ. To achieve "the faster the change, the greater the adjustment," you can use a piecewise function or a saturation function to avoid excessive adjustment.

[0046] Example formula, Δβ=η×arctan(μ×a_t) Update the weight: β_new=β_old+Δβ Weight normalization: Since α + β + λ = 1, after updating β, all weights need to be re-normalized, or the values of α and λ are adjusted accordingly to keep the sum to 1.

[0047] Step 5: The system records the recommendation effect after each weight adjustment (measured by user click rate, collection rate, purchase conversion rate, and other business indicators).

[0048] Through long-term A / B testing or reinforcement learning mechanism, the mapping function parameters in step 4 are optimized in reverse, so that the adjustment strategy can better fit the actual business growth.

[0049] The dynamic weight adjustment module periodically crawls social media data, analyzes the current popular elements, and dynamically adjusts the value of weight coefficient β accordingly. Social media is an important platform for fashion trends, and by crawling and analyzing the data, we can keep abreast of the current popular trends. When the popular trend changes faster, the adjustment range Δβ of β is larger, so that the weight of the popularity score in the matching degree calculation can quickly respond to the changes in fashion trends, ensuring the timeliness of the recommendation results.

[0050] Through interaction with social media data, popular trend information is obtained, and the weight coefficient β is adjusted according to the trend change degree, thereby affecting the calculation of the pattern matching degree. This allows the system to keep up with fashion trends and adjust the recommendation strategy in a timely manner, providing users with pattern recommendations that conform to the current popular trends and improving user satisfaction with the recommendation results.

[0051] The posture feature extraction module adopts a neural network model trained based on a federated learning framework. The model is trained locally on the user terminal or edge computing node, and only the update amount of the model parameters is encrypted and uploaded to the feature extraction server for aggregation. The original human body image data is retained locally.

[0052] A neural network model trained based on a federated learning framework is used for posture feature extraction. The model is trained locally on the user terminal or edge computing node, and only the update amount of the model parameters is encrypted and uploaded to the feature extraction server for aggregation. The original human body image data is retained locally. This training method can effectively protect the privacy of users, avoid the risk of leakage of original data during transmission and storage, and at the same time, utilize the computing resources of the user terminal or edge computing node for local training, reducing the computational burden of the server.

[0053] Model training is performed locally, and parameter updates are encrypted and uploaded to realize model aggregation and optimization without transmitting original data. This protects the privacy and security of users, improves the efficiency and security of data processing, and fully utilizes distributed computing resources.

[0054] The local training period E_l of the federated learning framework is inversely proportional to the number of users N, and is set as E_l=max(1, round(C / N)), where C is a constant and round is an integer function, to ensure that the system can efficiently protect privacy and quickly converge when the user scale expands.

[0055] When the number of users N increases, the local training period E_l decreases accordingly, which ensures that the system can still efficiently protect privacy and quickly converge when the user scale expands. Because the number of users increases means that more data is involved in model training, appropriately reducing the local training period can speed up the update of the model and improve the overall performance of the system. The system dynamically adjusts the local training period according to the change of the number of users to balance the efficiency of model training and the demand for privacy protection. The system can adapt to different sizes of user groups, ensuring that the system can maintain good performance and privacy protection effect when the number of users changes.

[0056] The system also includes a collision processing module for comparing the recommended list with a public pool of high-frequency use patterns, and introducing a penalty factor for patterns in the recommended list that have a matching degree above a threshold but also have a high frequency of use in the public pool, reducing their final ranking.

[0057] The module compares the recommended list with a public pool of high-frequency use patterns, and for patterns in the recommended list that have a matching degree above a threshold but also have a high frequency of use in the public pool, introduces a penalty factor to reduce their final ranking. This can avoid recommending too common patterns, increase the diversity and uniqueness of the recommendations, and meet the user's demand for personalized patterns. By comparing with the public pool and introducing the penalty factor, the ranking order of the patterns in the recommended list is adjusted. The degree of personalization of the recommended results is improved, and the "collision" phenomenon is avoided, providing users with more distinctive pattern choices.

[0058] The penalty factor γ of the collision processing module is calculated as follows: γ=1 / (1+k×F), where F is the frequency of use of the pattern in the public pool, and k is a suppression coefficient greater than 0. By adjusting k (0.1≤k≤0.5), the system's avoidance strength of the "collision" phenomenon can be controlled.

[0059] By adjusting the value of k (0.1≤k≤0.5), the system's avoidance strength of the "collision" phenomenon can be controlled. The larger the value of k, the greater the penalty for high-frequency use patterns, and the more common patterns can be avoided in recommendations; the smaller the value of k, the relatively weak penalty, and some more common but highly matching patterns are appropriately retained. The penalty factor is calculated according to the frequency of use of the pattern in the public pool and the suppression coefficient k, thereby affecting the ranking of the pattern in the recommended list. A flexible "collision" avoidance strategy is provided, which can adjust the value of k according to actual needs to balance the personalization and practicality of the recommended results.

[0060] The user terminal is also configured to receive feedback operations of the user on the recommended results; and the central processing server further comprises a feedback learning module configured to fine-tune the weight coefficients a and β corresponding to the user according to positive or negative feedback of the user, so as to realize iterative recommendation strategy of the user.

[0061] The user terminal receives feedback operations of the user on the recommended results, and the feedback learning module of the central processing server fine-tunes the weight coefficients a and β corresponding to the user according to positive or negative feedback of the user. Positive feedback indicates that the user is satisfied with the recommended results, and the relevant weight coefficients are appropriately increased, so that the system is more inclined to recommend similar style patterns in subsequent recommendations; negative feedback indicates that the user is not satisfied, and the weight coefficients are correspondingly reduced to adjust the recommendation strategy. In this way, iterative recommendation strategy of the user is realized, and the accuracy of the recommendation and the user satisfaction are continuously improved.

[0062] The system adjusts the weight coefficients according to the feedback information of the user, and optimizes the recommendation strategy. The user's demand and preference changes can be understood in real time, and the recommendation strategy is continuously iterated to provide the user with pattern recommendations that are more in line with the personalized needs of the user.

[0063] The system further comprises a virtual try-on interface generation module connected with the recommendation list generation module, configured to superimpose and render the recommended pattern ranked first through the augmented reality technology to the human body image uploaded by the user, to generate and output a virtual try-on effect picture.

[0064] The module is connected with the recommendation list generation module, and superimposes and renders the recommended pattern ranked first through the augmented reality technology to the human body image uploaded by the user, to generate and output a virtual try-on effect picture. The user can intuitively see the display effect of the pattern on his own body through the virtual try-on effect picture, and can experience the actual effect of the pattern in advance, thereby improving the purchase decision efficiency and satisfaction of the user.

[0065] The system uses the augmented reality technology to realize the fusion of the pattern and the human body image, and generates a virtual try-on effect. The system provides a more intuitive and convenient experience for the user, helps the user to better select a pattern suitable for himself, and enhances the interactivity and participation of the user and the system.

[0066] In embodiment 1, the pattern matching degree is calculated. When the user uploads a human body image, and the feature extraction server extracts the shoulder width of the user as 50 cm and the height as 170 cm, the ratio R_s of the shoulder width to the height is R_s = 50 / 170 ≈ 0.294. The threshold values T_low and T_high are set as 0.25 and 0.35 respectively, and since R_s < T_low, the layout sensitivity factor λ is set as the minimum value λ_min = 0.6.

[0067] The style analysis module identifies the user's dressing style label as "retro", and the style label of a certain pattern in the pattern database is also "retro". Through similarity calculation, the style score S_s=0.8 is obtained. Real-time trend data shows that the popularity score of the pattern is S_t=0.7. According to the user's body shape parameters, the layout of the pattern is highly consistent with the user's body shape, and the layout score S_l=0.9. Set the weight coefficients α=0.3, β=0.1 (initial value, which can be adjusted according to the dynamic weight adjustment module).

[0068] According to the pattern matching degree calculation algorithm Score=α×S_s+β×S_t+λ×S_l, we can get: Score=0.3×0.8+0.1×0.7+0.6×0.9 =0.24+0.07+0.54 =0.85 Example 2: Dynamic weight adjustment, the dynamic weight adjustment module extracts social media data every 24 hours for analysis. Suppose the current trend changes rapidly, and through analysis it is found that the heat of a certain type of pattern on social media has risen sharply. The initial weight coefficient β=0.1, according to the degree of change of the trend, the adjustment amplitude Δβ=0.2 is determined, then the adjusted weight coefficient β'=β+Δβ=0.1+0.2=0.3.

[0069] In subsequent pattern matching degree calculation, the adjusted weight coefficient β' will be used for calculation, so that the popularity score has a higher weight in the matching degree calculation, and can more timely reflect the current trend.

[0070] Example 3: Anti-collision processing, there is a pattern A in the recommendation list, its matching degree is 0.9, and after comparison with the public pool, it is found that the use frequency F=0.5 of pattern A in the public pool is 0.5, and the frequency threshold is set to 0.4, and the suppression coefficient k=0.3.

[0071] According to the punishment factor calculation method γ=1 / (1+k×F), we can get: γ=1 / (1+0.3×0.5) =1 / (1+0.15) =1 / 1.15≈0.87 Then the final matching degree of pattern A after anti-collision processing is 0.9×0.87≈0.783, and its ranking will be correspondingly reduced, avoiding recommending too common patterns.

[0072] Example 4: Feedback learning, the user gives positive feedback on pattern B in the recommendation list, the initial weight coefficient corresponding to this user is α = 0.3, β = 0.1. The feedback learning module increases α by 0.05 and β by 0.02 according to the positive feedback, and the adjusted weight coefficients are α' = 0.3 + 0.05 = 0.35 and β' = 0.1 + 0.02 = 0.12.

[0073] In subsequent recommendations, the system will be more inclined to recommend patterns similar in style to pattern B and with higher popularity, in order to better meet the user's personalized needs.

[0074] In summary, the intelligent pattern classification and recommendation system has the following advantages: Personalized recommendation accuracy: By considering multiple factors such as user body posture features, dressing style preferences, real-time trend, and pattern layout orientation, and using a scientific matching degree calculation algorithm, the system can provide highly personalized pattern recommendations to meet users' diverse needs.

[0075] Good privacy protection: The federated learning framework is used for posture feature extraction, and the original body image data is kept locally, only encrypted model parameter updates are uploaded, effectively protecting user privacy and security, and solving the privacy concerns of users in the data sharing process.

[0076] Follow the fashion trend: The dynamic weight adjustment module can periodically extract social media data, analyze current popular elements in a timely manner, and dynamically adjust the weight coefficients, so that the recommendation results can quickly respond to changes in fashion trends and provide users with pattern choices that meet current popular trends.

[0077] Avoid "collision" phenomenon: The anti-collision processing module reduces the recommendation ranking of high-frequency use patterns by comparing with the public pool and introducing a penalty factor, increases the diversity and uniqueness of the recommendation results, and meets the user's pursuit of personalized patterns.

[0078] Good user experience: The virtual try-on interface generation module uses augmented reality technology to provide users with virtual try-on effect pictures, allowing users to intuitively see the display effect of patterns on their bodies, improving user purchase decision-making efficiency and satisfaction; the feedback learning module can adjust the recommendation strategy in real time according to user feedback, continuously optimize the recommendation results, and further improve user experience.

[0079] It is to be understood that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting; it is not intended to exclude myriad other embodiments of the present application that other inventors can develop based on the description herein. A person of ordinary skill in the art will recognize that elements from the various embodiments can be combined to form additional embodiments. It is intended that the specification and examples be considered as exemplary only, with the true scope of the application being indicated by the following claims.

[0080] The foregoing is considered as illustrative only of the principles of the application. Further, since numerous modifications and changes will readily occur to those skilled in the art, it is not desired to limit the application to the exact construction and practice described. Accordingly, all such variations are intended to be included within the scope of the present application as defined in the following claims, along with the full scope of equivalents to which such claims are entitled. It is intended that the scope of the application herein disclosed should be determined by the claims and that such intent should be accorded with the rules of claim draftsmanship.

Claims

1. An intelligent ornament pattern classification and recommendation system, characterized in that, The application relates to a personalized pattern recommendation system, comprising: a user terminal for obtaining user inputted original data containing at least a human body image; a feature extraction server in communication connection with the user terminal, comprising: a posture feature extraction module for extracting human body posture key points of a user from the human body image and calculating body shape parameters including shoulder width and waist-hip ratio based on the key points; a style analysis module for analyzing user specified or identified dressing style labels from user historical images; a central processing server in communication connection with the feature extraction server and a pattern database respectively, comprising: a pattern matching degree calculation module for calculating the matching degree of each pattern in the pattern database with the user according to the body shape parameters and the dressing style labels; the calculation process introduces a layout sensitive factor lambda, and the value range of lambda is 0.6<=lambda<=1.2; the factor is dynamically adjusted based on the body shape parameters and is used for controlling the weight of the influence of pattern layout suggestion on the final matching degree; a recommendation list generation module for generating a personalized pattern recommendation list according to the matching degree sorting; the pattern database is used for storing pattern data and its associated attribute labels, and the attribute labels at least include pattern style, applicable part, layout orientation and popularity index. 2.The intelligent ornament pattern classification and recommendation system of claim 1, wherein, The pattern matching degree calculation module calculates the matching degree Score by executing the following algorithm: Score=alpha*S_s+beta*S_t+lambda*S_l wherein S_s is a style score based on the similarity between the user dressing style label and the pattern style label, S_t is a popularity score based on real-time trend data, S_l is a layout score based on the conformity between the body shape parameters and the pattern layout orientation; alpha and beta are configurable weight coefficients, and alpha+beta+lambda=1, wherein lambda is the layout sensitive factor. 3.The intelligent ornament pattern classification and recommendation system of claim 2, wherein, The layout sensitive factor lambda is dynamically set according to the ratio R_s of the user shoulder width to the height: when R_s is greater than a threshold T_high, lambda is set to a maximum value lambda_max (1.0<=lambda_max<=1.2) to enhance the avoidance effect of the inapplicability of horizontally expanded patterns to users with high shoulder width; when R_s is less than a threshold T_low, lambda is set to a minimum value lambda_min (0.6<=lambda_min<=0.8) to reduce the weight of layout suggestion. 4.The intelligent ornament pattern classification and recommendation system of claim 3, wherein, The central processing server further comprises a dynamic weight adjustment module for periodically grabbing social media data to analyze current popular elements and dynamically adjust the value of the weight coefficient beta accordingly; the faster the popular trend changes, the greater the adjustment range Delta beta of beta. 5.The intelligent ornament pattern classification and recommendation system of claim 1, wherein, The posture feature extraction module adopts a neural network model trained based on a federated learning framework; the model is locally trained on the user terminal or an edge computing node, and only the update amount of model parameters is uploaded to the feature extraction server for aggregation; the original human body image data is retained locally. 6.The intelligent ornament pattern classification and recommendation system of claim 5, wherein, The local training period E_l of the federated learning framework is negatively correlated with the number N of users, and is set as E_l=max(1, round(C / N)), wherein C is a constant, and round is an integer function, so that the system can still efficiently protect privacy and quickly converge when the scale of users expands. 7.The intelligent ornament pattern classification and recommendation system of claim 1, wherein, The system further comprises an anti-collision processing module configured to compare the recommended list with a public pool of high-frequency use patterns, and introduce a penalty factor for a pattern in the recommended list that has a matching degree higher than a threshold but also has a use frequency higher than a frequency threshold in the public pool, so as to reduce the final ranking of the pattern. 8.The intelligent ornament pattern classification and recommendation system of claim 7, wherein, The penalty factor γ of the anti-collision processing module is calculated as follows: γ = 1 / (1+kxF), where F is the use frequency of the pattern in the public pool, and k is a suppression coefficient greater than 0, which is used to control the avoidance strength of the system to the collision phenomenon by adjusting k, and k is selected as follows: 0.1≤k≤0.

5. 9.The intelligent ornament pattern classification and recommendation system of claim 1, wherein, The user terminal is further configured to receive feedback operations of the user on the recommended results, and the central processing server further comprises a feedback learning module configured to fine-tune the weight coefficients α and β corresponding to the user according to positive or negative feedback of the user, so as to realize iteration of a personalized recommendation strategy of the user. 10.The intelligent ornament pattern classification and recommendation system of claim 1, wherein, The system further comprises a virtual try-on interface generation module connected with the recommended list generation module, configured to superimpose and render the recommended pattern ranked first on a human body image uploaded by the user through augmented reality technology, to generate and output a virtual try-on effect picture.

Images