Large aircraft point cloud model self-supervision semantic segmentation method based on deep learning

Abstract

The invention discloses a large aircraft point cloud model self-supervision semantic segmentation method based on deep learning, and the method comprises the steps: collecting large-size aircraft point clouds from a plurality of stations through a laser tracker, and generating a plurality of groups of 3D point cloud data; constructing a global measurement field, performing initial registration onthe plurality of groups of 3D point cloud data based on the global measurement field, and performing fine registration on the initially registered point cloud data based on graph optimization to obtain a complete aircraft point cloud model; and carrying out self-supervised semantic segmentation on the complete aircraft point cloud model. According to the invention, an optical three-dimensional detection system measurement technology is utilized to process and analyze acquired 3D data, so that shape semantic segmentation on a large aircraft can be accurately and effectively carried out, the conception is reasonable, and automatic application can be realized in scenes such as aircraft safety inspection and the like in practice.

Description

Technical field [0001] The invention relates to the technical fields of deep learning, computer vision and graphics, in particular to a method for self-supervised semantic segmentation of a large airplane point cloud model based on deep learning. Background technique [0002] As the service life gradually increases, the corrosion resistance of a large number of aircraft decreases. The use of a large number of aircraft shows that the main failure mode of aircraft is structural failure caused by damage to structural parts such as cracks and corrosion. Serious structural failures must be grounded for strict and comprehensive maintenance. Therefore, in order to ensure the normal use and safety of aircraft in service, aircraft non-destructive testing is very important, and aircraft skin surface testing is the focus of aircraft non-destructive testing. The aircraft skin is fixed to the fuselage by rivets. As the aircraft is lifted and landed, the cabin skin is subjected to periodic su...

AI Technical Summary

Problems solved by technology

[0004] The purpose of the present invention is to provide a self-supervised semantic segmentation method for large aircraft point cloud models based on deep learning. The data collected by the optical three-dimensional detection system is used to complete the semantic segmentation of the large aircraft shape. It is a geometric deformation based on the global measurement field. The analysis method has the advantages of high precision, high efficiency, and small workload. It has a broad application prospect for large-scale aircraft maintenance. It not only solves the shortcomings of traditional horizontal measurement such as low efficiency and low precision, but also has theoretical application and practical engineering application value. also very big

Images