A Multi-Scale Cross-Attention Network for Intelligent Classification of Cultural and Creative Products

This study tackles the intelligent classification task for cultural-creative product images, where conventional approaches often fail to address the inherent diversity and semantic ambiguity of artistic designs. We propose a Multi-Scale Cross-Attention (MSCA) network that synergizes hierarchical feature extraction with adaptive attention learning. The framework employs parallel multi-scale convolutional layers to capture discriminative patterns across spatial granularities, while its novel cross-attention module dynamically fuses channel-wise relationships and spatial dependencies to prioritize semantically critical regions. Experimental results show promising performance: compared to mainstream baseline methods, the proposed approach achieves improvements of 5.48 % in accuracy (74.23 %), 9.00 % in recall (74.38 %), and 16.7 % in F1-score (66.04 %). These improvements demonstrate the effectiveness of our method in processing visual features of cultural and creative content. This work contributes to robust feature representation learning for artistic image analysis, providing practical value for cultural heritage digitization and creative industry applications.