Image Classification Of Second-hand Clothing Based On Convolutional Neural Networks

With the simplification of shopping methods and the promotion of fast fashion trends,clothing consumption and waste have significantly increased.Second-hand clothing trading is an important means of achieving "carbon neutrality".Accurate and efficient clothing image classification can greatly improve the effectiveness of clothing retrieval and recommendation,and promote second-hand clothing transactions.In recent years,with the rapid development of deep learning and the emergence of convolutional neural networks,the performance of clothing image classification has been greatly improved.In view of this situation,this article will conduct research on the algorithm of second-hand clothing image classification based on convolutional neural networks.Compared with new clothing,second-hand clothing images often have poor shooting conditions,with cluttered shooting scenes,complex backgrounds,and insufficient lighting.In addition,there are uncertain factors such as deformation,wrinkles,and occlusion in secondhand clothing,which often result in unsatisfactory classification results of existing second-hand clothing images and fail to meet the needs of the second-hand clothing classification business sector.To more effectively classify second-hand clothing,this article proposes a second-hand clothing image classification algorithm based on feature fusion and attention mechanism.ResNet50 is used as the basic clothing classification algorithm structure,and the algorithm model is optimized by using multi-stage convolutional layer feature fusion.When using convolutional neural networks to extract image features,the lower-layer features of the network have higher resolution and richer position and detail information,but they have weaker semantics and more noise due to less convolution processing.Conversely,the high-level features of the network have lower resolution and are difficult to capture the details of the image,but have stronger semantics.Fusing feature information extracted by multiple stage convolutional layers can effectively utilize more complementary advantages of different-level features of the network,thus enriching the feature information extracted by the model.At the same time,this article embeds channel and position attention modules in the model,which are used to learn the internal correlations between channels and the spatial dependencies of features,respectively.By adding the output results of these two attention modules,the feature representation can be further enhanced.For the collection of image data of second-hand clothing.On the one hand,network crawlers are used to collect online second-hand clothing images for training the network model.On the other hand,photographic equipment is used to shoot second-hand clothing and collect pictures for testing the classification effect of the network model.In addition,to further study the applicability of the proposed algorithm,expanded experiments were conducted using the Deep Fashion clothing dataset provided by the Multimedia Laboratory of the Chinese University of Hong Kong.According to the experimental results,the classification accuracy of the proposed algorithm on the self-built dataset and the Deep Fashion dataset has been improved by 1.95% and 1.76%,respectively,compared to the baseline classification algorithm,confirming the effectiveness and generality of the proposed algorithm in this thesis.