Research On Women’s Dressing Collocation Based On Deep Learning

The purpose of this paper is to decrease the gap between traditional garment industry,emerging science and technology,help communication among consumers,garment experts and intelligence,at the same time provide more convenient and accurate online shopping service to consumers,and develop the broader space for clothing e-commerce.In this article,deep learning knowledge and stylish skills would apply in attributing features of women’s image recognition and classification.In the end,the research provides reliable suggestions to people for the temperature and the occasions.Attributes of female clothes is a complicated system.This paper researches step by step.Firstly,a self-designed simple CNN network and the classical VGG16 model are used to classify the macro attribute categories of female clothes images,and has identified camisole or vest,shirt,blazer,trouser,skirt and jumpsuit in the women’s wear category.Both models have achieved good recognition results.Among them,the recognition effect of VGG16 model is slightly better than that of simple CNN network,but the complexity of it is significantly increased.Secondly,joint training is adopted to research the multi-task classification problems of attributes of female clothes based on the inception-v4 model and inception-resnet-v2 model.In that way,the female clothes images are described more comprehensively in machine language based on deep learning,and the detailed features,such as part length and collar design,are identified.The Inception-v4 model has achieved good results in training and prediction,and the accuracy of part length is better than that of collar design.Under the guidance of experts in the field of clothing,this paper sorts out a set of mapping principles of ladies garment attributes based on temperature and occasions.Combined with machine-recognized attribute labels and artificially added labels such as material and profile,the multi-layer perceptron model is used to solve the problem of multi-label classification of womenswear adapting with temperature and occasions.Finally,the collocation recommendation based on temperature and occasions is showed through the interactive interface.The matching results meet the basic needs of temperature and occasion,and achieve a beautiful and fashionable effect.And it realizes the communication between people and machines in the fashion field.