Classification And Retrieval Of Clothing Images Based On Multilayer Convolutional Neural Networks

With the continuous development of artificial intelligence,the emergence of e-commerce platforms has caused a boom in online clothing purchase.In order to better meet the needs of consumers,e-commerce platforms need to improve the classification information of clothing while increasing the variety of clothing.Clothing online trading volume is increasing day by day,and accurate classification of clothing is not only conducive to the rapid search of target clothing,but also to find the best cost performance by comparing other factors in relatively similar clothing styles.Therefore,it is very important to study a fast classification method with high accuracy for the massive online clothing data set.The traditional method of classifying massive clothing pictures requires manual labeling first,and then semantic classification according to the labels of pictures.This method consumes a lot of manpower,takes a long time,and has low efficiency.With the emergence of deep learning technology,the efficiency of image classification by neural network machine autonomous learning has been greatly improved.This topic is based on multilayer convolution Neural Network(Convolutional Neural Network,CNN)to the clothing image classification and retrieval done the following work:(1)Adopt the general crawler method to crawl the clothing pictures and videodata set in the e-commerce APP,and select representative categories for the main clothing categories of men's and women's.Image classification USES deep learning framework Caffe and common network models AlexNet,vgg-16 and GoogleNet,and adjusts the loss function and other parameters to see its maximum classification accuracy.(2)In this design,video classification of clothing is proposed for the first time,and the influence of different frame extraction methods on the classification results is compared,which is of great research significance for the classification of clothing images.(3)This topic designs and builds a web search system,and establishes a search interface.Js language and Html framework are used to write front-end web pages.The framework of back-end web pages is Flask.Cosine distance algorithm is used to calculate the distance between the input image vector and the vector in the feature database data set.The smaller the distance is,the higher the image similarity will be.The retrieval results return the first 20 images with high similarity.