Automated Matching of Ancient Bone Stick Fragments: Integrating Siamese Networks With Sequence Similarity Metrics for Multi-Feature Fusion

Ancient bone stick fragments excavated from the site of ancient Chang’an during the Western Han Dynasty present significant challenges for matching due to their sheer number, complex matching features, and the scarcity of reference samples. To address these challenges, this paper proposes a novel Automatic Matching Method for Bone Stick Fragments that integrates Siamese networks with Sequence Similarity Metric Algorithms. Our method reconstructs a Siamese feature extraction network by combining a Vision Transformer with a scaled-down version of EfficientNetV2, efficiently capturing both global and local features of the bone stick images. We further enhance the matching accuracy by coupling the Siamese network with the Fast Multiscale Derivative Dynamic Time Warping algorithm, which focuses on the intricate matching features on the image surface and Broken Edge Sequence Characteristics. Experimental results show that our approach achieves a Top-15 recall rate of 94.55%, significantly outperforming comparative algorithms. This advancement facilitates efficient and precise matching of bone stick fragments, offering a substantial improvement over existing methods.