MobileViGv2-A: A Lightweight Graph Convolutional Vision Model for Bronze Artifact Dating

This paper proposes a bronze dating method based on MobileViGv2-A, which provides solutions to the small sample problem and computing resource limitation in bronze dating research. This study designs a lightweight graph convolutional visual model suitable for bronze feature extraction, proposes an enhanced graph hybrid convolution block combined with a self-attention mechanism, and uses a variety of data enhancement and transfer learning methods to alleviate overfitting. Experimental results show that the proposed method achieves an accuracy of 86.5% in the bronze dating task, which is better than existing methods, such as MobileNet and MobileViG. And has lower computational complexity and stronger generalization ability. This study provides efficient and feasible technical support for the digital protection and research of cultural heritage such as bronzes.