Semantic segmentation and 3D reconstruction of CT images of bronze casting moulds and cores based on a deep learning method

This study presents an innovative approach combining semantic segmentation and 3D reconstruction to analyze CT images of bronze casting moulds and cores unearthed at the Taijiasi archaeological site. Leveraging the Swin-Unet deep learning architecture, the proposed MouldCTSegNet model achieves accurate material segmentation in challenging CT datasets characterized by low contrast and blurred boundaries. The segmented results are used to reconstruct precise 3D models of different material components using volume rendering. This method not only enhances the understanding of ancient combined-material moulding techniques but also provides an advanced tool for cultural heritage preservation, offering significant contributions to archaeology and related disciplines.