Laser scanning technology used in the intelligent installation of ship shafting has such advantages as non-contact, high-speed scanning and high-precision imaging. The laser point cloud data includes the size, position and direction information of space objects. Point cloud segmentation can greatly reduce the calculation scale of the data and improve the measurement efficiency of the relative pose of the butt flange.

In this paper, deep learning theory is used to study point cloud segmentation and obtain a point cloud dataset of flange parts. The PointNet model is used for training. Optimization strategies are formulated in three aspects, namely dropout regularization, learning rate attenuation and point cloud data enhancement, then tested on a ship shafting intelligent installation platform.

The convergence results of the model tend to be stable, with the accuracy of the training set reaching 0.88 and that of the verification set reaching 0.65. The flange point cloud segmentation experiment shows clear contour edges.

The results of this study show that the proposed method has good convergence and generalization performance, and can improve the efficiency of ship shafting intelligent installation.