Application study of PointNet++ and hybrid filtering based point cloud denoising and segmentation algorithms for Blisk

In the process of collecting 3D point cloud data of aviation blisk, it is inevitable to collect various noise points and background point cloud data, which will affect the subsequent 3D reconstruction process. Therefore, how to extract the effective blisk point cloud information and retain the accurate structural features as much as possible is an important step in the preprocessing of the overall blisk 3D reconstruction. To this end, this paper proposes a hybrid filtering algorithm to denoise the single-view blisk point cloud data to remove noise. At the same time, the octree downsampling method is used to simplify the blisk point cloud data, which is beneficial to speed up the subsequent point cloud segmentation efficiency. The PointNet++ model is used to train labeled data to accurately extract blisk point cloud data. The experimental results show that the method can effectively segment the effective information of the whole blisk point cloud model, and the segmentation accuracy reaches 97.4893%.