Intelligent personalized clothing matching method and device based on deep learning

Abstract

The invention discloses an intelligent personalized clothing matching method and device based on deep learning, and belongs to the comprehensive technical field of artificial intelligence, deep learning and computer vision technologies. According to the invention, personalized clothing matching suggestions are provided for the user. By analyzing factors such as body characteristics, personal preferences and on-line meteorological conditions of the day of the user, and combining dressing specifications of a large language model, a clothes matching scheme meeting styles and requirements of the user is generated. In the aspects of user interface design and algorithm optimization, the system realizes high user personalization and interactivity, and provides better fashion experience for the user. The method has the advantages that artificial intelligence fusion of the large language model, highly intelligent recognition of individual characteristics of the user and high user friendliness are achieved, and brand-new fashion matching experience is provided for the user.

Description

Technical Field

[0001] The present invention discloses an intelligent personalized clothing matching method and device based on deep learning, which belongs to the technical field of artificial intelligence, deep learning and computer vision technology. Background Art

[0002] Since the beginning of the 21st century, with the popularization of big data and powerful computing resources, deep learning, especially large language neural network models, has made significant progress in the past few years and has become one of the key technologies for solving complex problems and realizing artificial intelligence. The development of deep learning has benefited from algorithm optimization, model structure innovation and the application of large-scale data sets, bringing breakthrough progress to image recognition, natural language processing, speech recognition and other fields, and promoting the rapid development and widespread application of artificial intelligence technology. The training of large language models can combine network resources and text specifications to learn natural language rules and provide humanized suggestions and guidance. In the future, artificial intelligence will continue to develop, innovate and break through, and bring more intelligent and convenient applications and services to mankind.

[0003] People's demand for clothing matching and requirements for personal image are increasing with the rapid development of people's quality of life. With the improvement of people's living standards and the emphasis on personal image, fashion and clothing have become an indispensable part of modern society. However, for many people, choosing the right clothes is still a challenge, especially in the face of daily weather changes, workplace conditions, personal interests and body shape. There are prominent difficulties in choosing clothes. Therefore, a program that provides suitable clothing matching based on weather conditions, personal physical characteristics and workplace conditions has important practical value and market demand. Summary of the invention

[0004] In order to solve the problems existing in the prior art, the present invention provides a clothing matching device that uses deep learning, a large language model and combines real-time weather data to achieve dynamic personalized matching recommendations. It also combines the user's physical characteristics, such as height, body shape, occupation and other factors to provide users with clothing matching suggestions that are more in line with individual characteristics. According to the user's personal preferences and weather changes, the system can adjust the clothing matching suggestions in real time to ensure that users can enjoy the most suitable fashion experience under different weather conditions.

[0005] The technical solution adopted by the present invention is: an intelligent personalized clothing matching method based on deep learning, comprising the following steps:

[0006] S1. Personal data collection: collect user information: frontal face photo, personal clothing, body characteristics and city;

[0007] S2. The user inputs a clothing photo and uses the neural network to predict the clothing type and color.

[0008] S3. Set a specific temperature parameter S(i) for each garment according to its warmth retention, and establish a temperature zone based on the local minimum and maximum temperatures of the day extracted:

[0009]

[0010] Where n is the number of clothing items in this combination, is the minimum temperature, is the maximum temperature, t is the temperature parameter corresponding to the temperature;

[0011] S4. Determine the minimum number of clothes required to keep the user warm based on the lowest temperature, and based on this, select the number of clothes required to ensure aesthetics; and determine whether it is necessary to add clothes that have a warming property but do not have aesthetics; preset a set of clothes that meet specific conditions to calculate the clothing combination suitable for this temperature zone; and finally obtain the temperature clothing combination suitable for this temperature zone. And get their corresponding temperature priority j:

[0012]

[0013] Where n is the number of clothing items in this combination, is the minimum temperature, is the maximum temperature, t is the temperature parameter corresponding to the temperature; a is the warmth feedback constant, and its initial value is 0;

[0014] S5. The user's personal photo is used to predict his face shape and skin color through the neural network; then the corresponding BMI value is obtained based on the height and weight he inputs; the obtained face shape, skin color and BMI value are used as the input of the existing large language model Bert model, and the output is the clothing combination suitable for him; and the public clothing combinations are used as training sets for training to obtain a trained clothing combination model; then specific recommended combinations are given based on the user's face shape, skin color and BMI value through the trained clothing combination model ;

[0015] Finally, match clothes according to temperature Recommended pairing Priority is determined based on the degree of overlap of the collocation;

[0016]

[0017] Among them, b is the aesthetic feedback constant, and its initial value is 0;

[0018] Priority p= +j, the smaller the p, the higher the priority. Finally, the final clothing combination is output according to the priority.

[0019] 2. The intelligent personalized clothing matching method based on deep learning according to claim 1 is characterized in that:

[0020] The method provides a user scoring feedback module and sets a user warmth and aesthetic evaluation table; obtains the relevant data of warmth T and warmth feedback constant a, aesthetic B and aesthetic feedback constant b according to the values ​​selected by the user, and obtains a and b for the next priority screening according to the least squares.

[0021] An electronic device comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps of the above-mentioned intelligent personalized clothing matching method based on deep learning are implemented.

[0022] A non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-mentioned intelligent personalized clothing matching method based on deep learning.

[0023] Furthermore, the intelligent personalized clothing matching method based on deep learning includes the following steps:

[0024] S1. Personal data collection: collect user information, including: frontal face photo, personal clothing, body features and city;

[0025] S2. Set a specific temperature parameter S(i) (Table 1) for each garment according to its warmth retention, and establish a temperature zone based on the local minimum and maximum temperatures of the day, and calculate the clothing combination suitable for this temperature zone based on this; its expression is

[0026]

[0027] Where n is the number of clothing items in this combination, is the minimum temperature, is the maximum temperature, t is the temperature parameter corresponding to the temperature;

[0028] S3. Automatically identify clothing needs based on the clothing provided by the user, propose a dressing plan, and preset 6 clothing sets that meet specific situations for quick calculation. Then get the matching suitable for the temperature based on the temperature difference zone (the main purpose is to determine whether to add clothing with warmth properties).

[0029] surface

[0030]

[0031] S4. According to the obtained reasonable clothing matching that is suitable for the temperature ( ), combined with the big language model big data training samples and dress rules and regulations, Yang Bei systematically screened. At the same time, the built-in daily life experience is used to remove unreasonable items to avoid the situation where some specific clothes do not match. Through the personal photos input by the user, his face shape and skin color are predicted by the neural network, and then the corresponding BMI is obtained according to his input height and weight; through the three quantities obtained, the existing text and photos about clothing matching are trained as pre-training training sets using the big language model, and the trained big language model is built into the system, so that it can give specific relevant models according to the user's physical characteristics (face shape, skin color, BMI) and weather or according to the user's needs, and the original suitable clothing matching is screened to obtain the most suitable clothing matching for the user. According to the reasonable clothing matching that is adapted to the temperature, combined with the big language model big data training samples and the dress rules and regulations samples, it is systematically screened, and at the same time, the built-in daily life experience is used to remove unreasonable items to avoid the situation where some specific clothes do not match.

[0032] S5. On this basis, the obtained matching combinations are prioritized according to the following formula based on user characteristics. The smaller j is, the higher the priority is. The initial values ​​are a1=0, a2=0.1, and a2 is the preset temperature feedback constant when matching for the second time. The specific value of a is continuously optimized based on user feedback. Finally, a personalized clothing matching that meets the weather conditions and fits the user's personal characteristics is obtained.

[0033]

[0034] The entire matching system and model provide user scoring feedback function, optimize the algorithm based on user feedback data, and finally build in a user warmth and aesthetic evaluation table (for example, Table 2). According to the value selected by the user, the relevant data of T and a can be obtained, and the zero point of T, i.e. a at the next priority screening, can be obtained through least squares, so that the system can continuously adapt to users.

[0035] Table 2 User evaluation table

[0036] Warmth (T) Cold (-4) -3 -2 -1 0 (Appropriate) 1 2 3 Heat (4) Score Beautiful (B) 0 1 2 3 4 5 6 7 8 Score

[0037] The beneficial effects of the present invention are:

[0038] The present invention provides users with personalized clothing matching suggestions. By analyzing factors such as the user's physical characteristics, preferences, and weather conditions, a clothing matching plan that meets the user's style and needs can be generated. In terms of user interface design and algorithm optimization, the system achieves a high degree of user personalization and interactivity, providing users with a better fashion experience.

[0039] The invention can continuously optimize the algorithm according to user feedback data and provide clothing recommendations related to the user's physical characteristics and meteorological conditions. By combining with user behavior analysis and clothing matching association models, accurate personalized clothing matching suggestions are achieved to meet the user's needs for fashion and comfort. The advantages of the invention are that it is highly intelligent and user-friendly, providing users with a new fashion matching experience.

[0040] This system takes into account the requirements of dress code, personality characteristics, occupational characteristics, weather environment, and body shape. By controlling the matching of clothing, it ensures the comprehensive effect of multiple goals such as warmth, personality, and workplace. It ensures beauty by matching the color of clothing according to the conditions of physical characteristics. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 It is a flow chart of the intelligent personalized clothing matching method based on deep learning. DETAILED DESCRIPTION

[0042] In order to enable those skilled in the art to better understand the technical solution of the present invention, the present invention is described in detail below in conjunction with specific implementation steps, but it is not intended to limit the present invention. For the steps described herein, if there is no necessity for a causal relationship between each other, the order in which they are described as examples herein should not be regarded as a limitation, and those skilled in the art should know that the order can be adjusted as long as the logic between them is not destroyed and the entire process cannot be implemented.

[0043] A clothing matching device uses deep learning technology combined with real-time weather data to achieve dynamic personalized matching recommendations. It also combines the user's physical characteristics, such as height, body shape, etc., to provide users with clothing matching suggestions that are more in line with individual characteristics. According to the user's personal preferences and weather changes, the system can adjust clothing matching suggestions in real time to ensure that users can enjoy the most suitable fashion experience under different weather conditions.

[0044] It can provide suitable clothing matching and tips based on the user's personal frontal face photo, personal clothing, personal physical characteristics and the city they live in. The application obtains the local climate based on the city, gives weather-appropriate clothing based on the relationship between weather and clothing, and then obtains the final clothing matching based on the user's physical characteristics.

[0045] Through the clothing photos input by the user and the prediction of the pre-trained neural network, his clothing type and color are obtained. According to the warmth retention degree of the clothing, a specific temperature parameter S(i) is set for each clothing, and a temperature zone is established based on the extracted local minimum and maximum temperatures of the day. First, according to the minimum temperature, it is determined how many pieces of clothing are needed to ensure the user's warmth retention needs. Based on this, it is screened how many pieces of clothing should be needed to ensure beauty (externally visible) (for example, 1 piece: dress, two pieces: T-shirt + outer pants, three pieces: outer coat + inner wear + outer pants) and it is determined whether it is necessary to add clothing with warmth retention properties but not aesthetic needs, what kind of clothing to add to the user, store these contents and finally output them to the user through prompts. Six clothing sets that meet specific situations are preset for quick calculation of clothing combinations suitable for this temperature zone. Finally, the clothing combination suitable for this temperature zone is obtained ( ) and obtain their corresponding temperature priorities (j). The temperature band expression is

[0046]

[0047]

[0048] Where n is the number of clothing items in this combination, is the minimum temperature, is the maximum temperature, and t is the temperature parameter corresponding to the temperature.

[0049] Through the personal photo input by the user, the pre-trained neural network predicts his face shape and skin color, and then obtains the corresponding BMI based on his input height and weight; through the three quantities obtained, the existing large language model Bert model is used to adjust them, and the existing text about clothing matching is used as a pre-training training set to obtain a large language model about clothing matching, and adjust the input and output (for example, the input is the user's face shape, skin color, BMI, and the output is the clothing matching that suits him)

[0050] It can also provide specific matching according to the user's physical characteristics (face shape, skin color, BMI) and weather or according to the user's needs ( ), these combinations are used as the recommended combinations output by the system. According to its recommendation, the appropriate clothing combinations originally selected based on the warmth characteristics ( ) to filter out the most suitable clothing combination for the user ( ). The specific screening is based on how much overlap it has with the collocation given by the large language model or judging which clothes have the highest frequency in the collocation given by the large language model to determine the priority. related The smaller the value, the higher the priority.

[0051]

[0052] Priority (p=i+j), the smaller p is, the higher the priority is. Finally, the final clothing combination is output based on the priority.

[0053] Through user feedback, we continuously optimize it to make it more suitable for users. The specific operation is to prioritize the obtained matching combinations according to the following formula based on user characteristics. The smaller p is, the higher the priority is. The initial value of a is a1, which is 0, and a2=0.1. a2 is the preset temperature feedback constant for the second matching. The specific value of a is continuously optimized based on user feedback. The initial value of b is b1, which is 0, and b2=0.1. b2 is the preset aesthetic feedback constant for the second matching. The specific value of b is continuously optimized based on user feedback. Finally, a personalized clothing matching that meets weather conditions and fits the user's personal characteristics is obtained.

[0054] The entire matching system and model provide user scoring feedback function, optimize the algorithm according to user feedback data, and finally build in the user warmth and beauty evaluation table (for example, Table 2). According to the values ​​selected by the user, the relevant data of T and a, B and b can be obtained. According to the least squares, the zero point of T can be obtained, that is, a at the next priority screening, and b is the same, so that the system can continuously adapt to users.

[0055] Table 2 User evaluation table

[0056] Warmth (T) Cold (-4) -3 -2 -1 0 (Appropriate) 1 2 3 Heat (4) Score Beautiful (B) 0 1 2 3 4 5 6 7 8 Score

[0057] When using it, you must first preset the corresponding age, height, weight, and city. Then enter the corresponding personal photo. You only need one personal photo and the photo should be a frontal photo of the face in sufficient light. When entering clothing photos, please pay attention to first sort the clothing photos into a folder and then select all when entering. Failure to select clothing photos in this way may cause program errors.

Claims

1. An intelligent personalized clothing matching method based on deep learning, characterized in that: The following steps are involved: S1. Personal data collection: collect user information: frontal face photo, personal clothing, body characteristics and city; S2. The user inputs a clothing photo and uses the neural network to predict the clothing type and color. S3. Set a specific temperature parameter S(i) for each garment according to its warmth retention, and establish a temperature zone based on the local minimum and maximum temperatures of the day extracted: ； Where n is the number of clothing items in this combination, is the minimum temperature, is the maximum temperature, t is the temperature parameter corresponding to the temperature; S4. Determine the minimum number of clothes required to keep the user warm based on the lowest temperature, and based on this, select the number of clothes required to ensure aesthetics; and determine whether it is necessary to add clothes that have a warming property but do not have aesthetics; preset a set of clothes that meet specific conditions to calculate the clothing combination suitable for this temperature zone; and finally obtain the temperature clothing combination suitable for this temperature zone. And get their corresponding temperature priority j: ； Where n is the number of clothing items in this combination, is the minimum temperature, is the maximum temperature, t is the temperature parameter corresponding to the temperature; a is the warmth feedback constant, and the initial value a1 is 0; S5. The user's personal photo is used to predict his face shape and skin color through the neural network; then the corresponding BMI value is obtained based on the height and weight he inputs; the obtained face shape, skin color and BMI value are used as the input of the existing large language model Bert model, and the output is the clothing combination suitable for him; and the public clothing combinations are used as training sets for training to obtain a trained clothing combination model; then specific recommended combinations are given based on the user's face shape, skin color and BMI value through the trained clothing combination model ; Finally, match clothes according to temperature Recommended pairing Priority is determined based on the degree of overlap of the collocation; ； Among them, b is the aesthetic feedback constant, and its initial value is 0; Priority p= +j, the smaller the p, the higher the priority. Finally, the final clothing combination is output according to the priority.

2. The intelligent personalized clothing matching method based on deep learning according to claim 1 is characterized in that: The method provides a user scoring feedback module and sets a user warmth and aesthetic evaluation table; obtains the relevant data of warmth T and warmth feedback constant a, aesthetic B and aesthetic feedback constant b according to the numerical value selected by the user, and obtains the warmth feedback constant a and aesthetic feedback constant b for the next priority screening according to the least squares.

3. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps of the intelligent personalized clothing matching method based on deep learning described in claim 1 or 2 are implemented.

4. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the deep learning-based intelligent personalized clothing matching method described in claim 1 or 2 are implemented.

Images