Research On Clothing Classification Algorithm Based On Deep Learning

With the widespread popularity of e-commerce platforms,online shopping has gradually become a major part of people’s lives,and clothing is also one of the main commodities of online shopping.In order to make it convenient for users to buy clothes on the platform,merchants need to label the categories of clothes and upload them to the e-commerce platform,so clothing classification is a very important process.Marking a large number of clothing categories manually is not only a heavy workload but also subjective,which is difficult to meet the needs of users.Therefore,automatic labeling by computer has become the main way of classification.At present,the deep learning method is mainly used to study the clothing classification algorithm,but the accuracy of fine-grained clothing classification is still not high due to many categories of clothing,and the shooting of many clothing pictures on the platform is affected by various factors such as illumination,which leads to the inaccurate classification result.These problems bring many difficulties to the research of clothing classification algorithm.In order to improve the accuracy of clothing classification,this paper studies the clothing classification algorithm based on deep learning.In this paper,a Res Net50-DS clothing classification algorithm is proposed.Because there are multiple subcategories of collar and length,in order to avoid more errors when learning the category characteristics of various clothing,Res Net50 is used as the basic network to introduce the attention mechanism SE-Net into the residual network,which can remove redundant information,enhance the perception of the length and collar area of clothing in the clothing image,focus on the key categories to extract features on the collar and length,and suppress the interference of other irrelevant features.However,the network is prone to over fitting during training,which will cause some generalization errors,which will lead to the accuracy of clothing classification and affect the performance of the model.Therefore,combined with regularization,multiple Dropout are added to the connected MLP multi-classification task to improve the generalization ability of the multi-classification model,reduce the error of model generalization,and improve the network performance,thereby improving the accuracy and accuracy of collar and length,and optimizing the image classification algorithm in the clothing field.