In order to improve the quality of 3D point cloud fusion of ceramic artifacts, a new 3D point cloud fusion method is proposed by using graph neural network. Firstly, the image data of ceramic artifacts are collected; secondly, the point cloud is sampled and reconstructed by using the uniform sampling method; on this basis, the multi-scale features of the artifacts are extracted based on the graph neural network; Finally, the artifacts are fused into the 3D point cloud, and the fusion results are post-processed and optimized. The test results show that after the proposed method is applied, the peak signal-to-noise ratio of the image is higher, and the original image and the fused image are closer to each other in the corresponding pixels, which can more accurately restore the texture details, edge sharpness and geometry of the ceramic artifacts.