Research And Implementation Of Clothing Image Retrieval Method Based On Deep Learning

With the rapid development of e-commerce,people have gradually developed the habit of online shopping,and the proportion of clothing online transactions has also continued to increase,resulting in a rapid increase in the amount of clothing image data on the Internet.How to accurately retrieve the clothing images that people need from a large number of clothing image databases on the Internet has gradually become a hot research topic.The traditional text-based clothing image retrieval method requires manual labeling,and has shortcomings such as unclear description and strong subjectivity,which makes the retrieval results inaccurate.Traditional content-based clothing retrieval methods have problems such as background interference and occlusion,and the extraction of clothing features has certain limitations.In recent years,in the field of computer vision,deep learning has solved many problems that cannot be solved by traditional methods,and deep learning has been widely used.Therefore,this paper will combine deep learning technology to carry out research on clothing image retrieval from two different perspectives.This paper studies clothing image retrieval methods based on deep learning technology,studies and proposes two methods of clothing image retrieval based on attention mechanism and clothing image retrieval based on image-text combination,and designs and implements a combined imagetext clothing retrieval system.The specific work is as follows:(1)A clothing image retrieval method based on attention mechanism is proposed.Since clothing images inevitably contain some interference factors,such as light,background and occlusion,these interference factors will be extracted together when using convolutional neural network for feature extraction,which will affect the retrieval accuracy.In order to reduce the influence of these interference factors,an attention mechanism-based clothing image retrieval method is proposed.A lightweight general attention module is added to the convolutional neural network.By sequentially extracting feature maps from two independent dimensions of channel and space,it improves the attention of the clothing area in the process of feature extraction,and suppresses interference areas such as background attention,so as to improve the expression ability of image features.Comparing experiments with the network without attention mechanism on the Deepfashion dataset,the experimental results show that the method based on the attention mechanism proposed in this paper can effectively improve the accuracy of clothing retrieval.(2)A clothing image retrieval method based on image-text combination is proposed.In order to meet the needs of users to modify a certain attribute in the retrieved clothing images,it is realized through joint retrieval of images and texts.This paper proposes a multimodal fusion query method,which uses clothing images and modified attribute texts for retrieval,combines text and image descriptions,and replaces some image features with text description features.In order to achieve this retrieval task,a novel image-text feature fusion method is proposed,which is a multimodal combined feature used to describe text and images by combining gating and residuals.Finally,a comparative experiment is carried out on the Fashion-200 K dataset with other retrieval methods to verify the superiority of the proposed multimodal fusion retrieval method.(3)Design and implementation of clothing image retrieval system.By using the combined image and text retrieval method proposed in this paper,a web-based clothing image retrieval system is designed.The whole system is implemented using the technology of separating the front and back ends.The front part uses the Vue framework and the back end uses the Spring Boot framework.Users can upload clothing images and input modified text in the system,and then they can retrieve clothing images that meet the requirements in the data set.