Application research of multi-category commodity classification algorithm based on vision transformer

With the rapid development of technology, people’s shopping methods have also undergone earth-shaking changes. In order to quickly respond to and meet customers’ increasingly diverse online shopping needs, multi-category product classification technology has therefore become a crucial technology to improve user shopping experience and e-commerce platform operation efficiency, quickly and accurately classifying products from massive amounts of product data. Identifying products that users like has become a pain point that e-commerce platform systems urgently need to solve. In recent years, Vision Transformer (ViT) has demonstrated excellent performance in the field of image classification. This paper aims to propose a multi-category product classification algorithm based on ViT to improve the classification accuracy on large-scale product image data sets. Compared with traditional convolutional neural networks (CNN), ViT uses the self-attention mechanism in the Transformer architecture to better capture long-distance dependencies in images, and the recognition effect on large-scale data sets is particularly significant. In order to fully verify the superiority of the proposed method, we constructed a data set containing tens of thousands of product images of different categories, and conducted full experimental analysis based on this data set. Experimental results show that the model achieves significantly better performance than traditional methods on the standard test set.