Research on Clothing Style Recognition and Recommendation System based on Improved YOLOv3

Clothing style recognition is an important application of computer vision in recent years. With the rapid development of deep learning technology, image retrieval technology based on convolutional neural networks has made significant breakthroughs in the field of clothing. With the increase in the number and variety of clothing, people need to spend more time selecting suitable clothing products, and a good recommendation system is particularly important. Firstly, in order to fully explore the multi-scale features of clothing images from global to local, and make the model pay more attention to the clothing itself, a multi-scale clothing image landmark area positioning algorithm based on YOLOv3 was proposed, which can extract global features, main features and local features more accurately. Next, in order to make clothing image retrieval truly focus on the retrieved clothing and accurately extract clothing style features, the Delete Common Features (DCF) algorithm and Combine Feature Descriptors (CFD) algorithm were applied to different network convolutional layers. Finally, a clothing style feature extraction model (MCSF) based on multi-scale convolutional neural networks was proposed to achieve accurate clothing style recommendation. The experimental results show that the proposed model has higher accuracy and accuracy compared to other network models.