Enhancing Fashion Image Classification in E-Commerce Information Systems: An Integrated Approach

Abstract

Fashion image classification presents a challenging task due to the wide variety of clothing styles and their inherent semantic relationships. While convolutional neural networks (CNNs) such as ResNet50 excel in extracting spatial features,they often fall short in capturing the relational dependencies between fashion images, which are critical in this domain. This study proposes an innovative approach that integrates ResNet50 for spatial feature extraction with a graph convolutional network (GCN) to model relational dependencies using a dynamic graph structure. Evaluated on the DeepFashion dataset [1], the proposed method demonstrates superior performance compared to standalone CNN models, achieving an accuracy improvement of 10%. These findings highlight the potential of combining spatial and relational learning to advance fashion image classification.