Clothing Image Classification And Retrieval Based On Multitask Convolutional Neural Network

With the rapid development of Internet e-commerce,the amount of online clothing transactions is increasing day by day,and the importance of clothing images in transactions has also increased.However,there are many types of clothing and the classification standards are different.Consumers and e-commerce sellers are difficult to unify the description of the categories of clothing,which may lead to a poor shopping experience.Therefore,how to effectively identify clothing images is a very significant issue.In recent years,the deep convolutional neural network has an excellent performance in the field of computer vision.This article aims to use relevant technologies to study clothing image recognition,and the clothing image recognition atmosphere three tasks of detection,classification,and retrieval.And completed the following work:First,in order to improve the computational efficiency and training convergence speed of deep convolutional neural networks,a more lightweight deep convolutional neural network(Lighten-VGGNet)is proposed by analyzing classical models and tuning them.And it is used as a backbone network to support the task of detecting,classifying and retrieving clothing images.Secondly,this article aims at the characteristics of clothing image data,improves two clothing detection methods based on deep learning,and chooses to use in response to different scenarios.Then,aiming at the problem of too large granularity of clothing categories in clothing classification,this paper proposes a new hierarchical labeling strategy to refine clothing classification.Based on this strategy,a multitasking classification method is used to classify the garment image data after layered labeling.Without significantly increasing the training cost of the network model,the accuracy of the clothing classification is improved,and features extracted by the classification network are made available.Ability to express multiple garment category attributes.Finally,in the clothing image retrieval stage,based on the features extracted from the hierarchical multi-tasking classification network,garment image retrieval is performed using the spatial distance method.This article provides a basic support for garment image recognition application scenes by researching clothing image detection,classification and retrieval methods.