The number of fragments and the variety of primitive cultural relics unearthed in archaeology, especially the mixed fragments of several dynasties unearthed in Qinglong town, Shanghai, pose a great challenge to the manual splicing. The traditional manual comparison method is easy to cause the second damage to the cultural relics. In this paper, the edge feature is extracted based on removing the noise of point cloud, a bilateral filtering point cloud denoising algorithm based on salient features is proposed. By changing the step size and field of view, the Improved Artificial Fish Swarm Algorithm is used to get the matching strategy, and the point cloud is used to reconstruct 3D model by the Dual Quaternion Transformation method. The pairing of fragments and virtual reconstruction can effectively avoid the secondary damage of cultural relic fragments. It provides a feasible artificial intelligence solution for the protection and restoration of similar archaeological excavations.