Search

CN-122022942-A - Functional clothing optimization system and method

CN122022942ACN 122022942 ACN122022942 ACN 122022942ACN-122022942-A

Abstract

The invention provides a functional clothing optimization system and method, and belongs to the field of clothing design. The system comprises a user input module, a data acquisition module, a core database, an intelligent collocation engine, a visual output module and a feedback learning module, wherein the user input module is used for receiving collocation requests initiated by users, analyzing the collocation requests, extracting active scenes, active time and active place keywords, the data acquisition module is used for calling a third party API to acquire real-time environment data according to the active places and the active time extracted by the user input module, the core database is used for storing structured data of clothes and accessories, the intelligent collocation engine is used for generating and arranging collocation schemes through a multilevel rule query and sequencing algorithm, the visual output module is used for visually displaying the collocation schemes in a virtual try-through image mode, and the feedback learning module is used for collecting scores and feedback of recommended schemes by users and experts and optimizing the recommended models based on the feedback data. The invention solves the problem that the traditional fashion recommendation system has a pain point of functional deficiency in specific scenes such as professional sports, outdoors and the like.

Inventors

  • SUN CHAO
  • LI XIN

Assignees

  • 中华女子学院

Dates

Publication Date
20260512
Application Date
20251225

Claims (10)

  1. 1. A functional apparel intelligent recommendation system, the system comprising: the user input module is used for receiving collocation requests initiated by users in natural language, analyzing the collocation requests, extracting keywords of activity scenes, activity time and activity places, and managing personal information of the users; The data acquisition module is used for calling a third party API to acquire real-time environment data according to the activity place and the activity time extracted by the user input module; The core database is used for storing the structured data of the clothing and the accessories and labeling functional attributes, layer attributes, application scene attributes, aesthetic attributes and basic information for each article; the intelligent collocation engine is used for generating a collocation scheme through a multilevel rule query and sequencing algorithm based on user input information, real-time environment data and data of a core database; the visual output module is used for intuitively displaying the collocation scheme in a virtual try-on image mode; And the feedback learning module is used for collecting scores and feedback of the user and the expert on the recommended scheme and optimizing the recommended model based on feedback data.
  2. 2. The functional apparel intelligent recommendation system of claim 1, wherein the intelligent collocation engine performs operations comprising: The rule inquiry is carried out, a preset wearing rule table is inquired according to the movement intensity and the temperature interval, and the required clothing layer system and the functional requirements of each layer are determined; the clothing retrieval is carried out, and the clothing single articles meeting the conditions are retrieved from a core database based on the functional requirements; the multiple sets of solutions are weighted scored and ranked based on the user preference matching.
  3. 3. The functional apparel intelligent recommendation system of claim 2, wherein the feedback learning module performs operations comprising: the method comprises the steps of constructing a positive sample pool and a negative sample pool according to user scores and expert scores, training a classification model based on sample data, dynamically adjusting weight parameters in a sequencing algorithm, analyzing expert evaluation texts through natural language processing, and updating a collocation principle knowledge base.
  4. 4. The intelligent functional clothing recommendation system according to claim 3, wherein the visual output module comprises an image synthesis sub-module for synthesizing clothing pictures in a collocation scheme into a virtual try-on effect map based on a user body size and a virtual portrait model.
  5. 5. The functional apparel intelligent recommendation system of any of claims 2-4 wherein the multi-level rule query comprises: First level, querying an activity-layer system rule table; A second stage of querying a rule table of 'temperature-layer system-attribute' based on the layer system and the temperature; And thirdly, executing database query, converting the rule into a WHERE condition of SQL, and querying the clothing table.
  6. 6. The functional apparel intelligent recommendation system of claim 5 wherein the ranking algorithm uses a weighted scoring model to calculate a preference score for each set of schemes using the formula: score = color preference match ×w1+ style preference match ×w2+ brand preference match ×w3 Wherein, W1, W2, W3 are weights, the initial value is set by expert, and the later stage is dynamically adjusted by the feedback learning module.
  7. 7. An intelligent recommendation method for functional clothing is characterized by comprising the following steps: S1, receiving natural language input of a user, analyzing and extracting information of an activity scene, time and place, and acquiring real-time weather data; S2, determining clothing layer systems and functional requirements based on the analyzed activity scenes and weather data, and searching single products meeting the conditions from a clothing database by inquiring a rule table; S3, combining the retrieved single products into a plurality of complete matching schemes, and sorting the complete matching schemes based on user preference; s4, outputting the collocation scheme in a visual form, and displaying the virtual try-on effect; And S5, collecting scoring and behavior data of the scheme by the user and the expert, optimizing the recommendation model and updating the knowledge base.
  8. 8. The intelligent recommendation method for functional clothing according to claim 7, wherein the step S1 comprises intention recognition, exercise intensity classification, temperature interval mapping and outputting structured JSON data through a large language model.
  9. 9. The intelligent recommendation method for functional apparel according to claim 8, wherein the operation of step S3 includes: S301, determining the requirements of clothing layer systems and functional attributes through multi-level rule inquiry; S302, optimizing color and style coordination based on a collocation principle knowledge base; and S303, calculating the user preference matching degree by adopting a weighted scoring model, and sequencing the schemes.
  10. 10. The intelligent recommendation method for functional apparel according to claim 9, wherein the operation of step S5 includes: s501, constructing a positive and negative sample pool based on dominant score and indirect behavior data; s502, training a classification model by using a machine learning algorithm, and optimizing weight distribution; S503, dynamically adjusting the sorting weights according to the model output, and updating the collocation principle knowledge base.

Description

Functional clothing optimization system and method Technical Field The invention relates to the field of clothing design, in particular to a functional clothing optimization system and method. Background With the deep fusion of artificial intelligence and fashion industry, intelligent clothing collocation recommendation systems have become a hotspot for research and application. Such systems aim to provide personalized, scenic dressing advice to users by technical means to reduce decision cost and enhance wearing experience. Currently, research and commercial practice in this area have mainly progressed along several technological paths, but there are certain limitations. The first type of scheme is based on leading edge generated Artificial Intelligence (AIGC). For example, in the prior art (such as CN120316250 a), a Large Language Model (LLM) is used to analyze the natural language requirement of the user, and a virtual try-on image is directly generated by coupling with a text-to-image model, so that the end-to-end output from text to an image is realized, and the user experience is visual. However, this solution has the significant drawback that, firstly, it generates a virtual clothing picture, which is not a real commodity, and cannot be directly associated for purchase. Second, large language models and meridional graph models are essentially "probabilistic generative models", the primary goal of which is to generate logically or visually reasonable, aesthetically pleasing content in language, rather than to ensure functional reliability. When facing professional scenes such as skiing, mountain climbing and the like, the system is very easy to ignore core function attributes (such as waterproof level, air permeability index and warmth retention level) of the clothing, and can recommend matching with vision fashion but with function deficiency, so that potential safety hazards exist. In addition, the "illusion" problem inherent to the generated AI and output randomness result in unstable recommended quality, lack of consistency and reliability. Meanwhile, the deployment and operation of the large model require extremely high computing resources, so that the system is high in cost and long in response delay, and real-time interaction and large-scale business are difficult to realize. This approach is typically focused on one-time generation, does not build an efficient continuous feedback optimization mechanism, and the enormous cost of fine-tuning large models also hinders deep personalized evolution for individual users. The second category focuses on matching recommendations based on machine vision and attribute analysis. The prior art (CN 119003812A) recommends garment properties as bridges by constructing a multi-level mapping model of 'occasion-property-image'. This approach avoids the difficulty of directly learning complex mappings, but is essentially an offline matching system that relies on static annotation data. The method can not actively acquire dynamic information such as real-time weather, popular trends and the like, and has insufficient flexibility. Although user ratings are mentioned, their core models are not learned from the user's personal history data, making deep personalized recommendations difficult. More importantly, the 'clothing attribute' defined by the method focuses on vision and shape characteristics such as styles, sleeve lengths and the like, and the consideration of key functional attributes such as water resistance, ventilation, warmth retention and the like is seriously lacking, so that the functional requirements of professional sports and outdoor scenes cannot be met completely. A third category of solutions is directed to providing closed-loop services from recommendation to access in combination with a user entity wardrobe. Representative technologies (such as CN 106485515A) can screen clothes from a user's private wardrobe and match the clothes according to rules by acquiring weather information, and even can indicate clothes storage positions, so that the practicality is strong. The method calculates preference weight through user history accepted data, and realizes lightweight personalized learning. However, the recommendation range of the system is strictly limited by the existing clothes stock of the user, the clothes (such as professional skis) which are not owned by the user and have specific functions cannot be recommended, and the application scene is limited. The matching rule base is statically preset, so that complex fashion concepts are difficult to understand or rapid change trend is adapted, and the recommendation result can be monotonously outdated. Furthermore, the system focuses mainly on weather adaptation and historical preferences, and does not take into account deep functional needs by wearing shapes or meeting specific activities (such as high intensity sports), which are not professional enough. The current state of academic research als