基于改进PointNet++的船体分段合拢面构件智能识别算法研究
Intelligent recognition algorithm for hull segment closure surface components based on improved PointNet++
李瑞 1赵怡荣 2霍世霖 2汪骥 1史卫东3
作者信息
- 1. 大连理工大学 船舶工程学院,辽宁 大连 116024;大连市舰船先进制造技术重点实验室,辽宁 大连 116024
- 2. 大连理工大学 船舶工程学院,辽宁 大连 116024
- 3. 大连船舶重工集团有限公司,辽宁 大连 116011
- 折叠
摘要
[目的]三维扫描仪获得的船体分段合拢面点云数据,具有精度高、数据量大的优势,能够很好地反映分段合拢面的建造状况.由于现有的PointNet++网络无法处理大容量点云数据,因此提出一种基于改进PointNet++的船体分段合拢面构件智能识别算法,实现针对大容量船体分段合拢面点云数据构件的智能识别.[方法]基于超体素生长理论对船体分段合拢面点云数据进行分割及简化,构建船体分段合拢面点云数据集,并使用该数据集训练基于深度学习理论改进的PointNet++网络.[结果]网络模型在船体分段合拢面点云数据训练集和测试集上的收敛结果趋于稳定,在测试集上识别准确率达到 90.012%.[结论]该方法具有良好的识别能力,能够完成船体分段合拢面构件的智能识别.
Abstract
[Objectives]The point cloud data of hull segment closure obtained by a 3D scanner has such ad-vantages as high precision and large data volume,and can accurately reflect the construction status of segment closure.Since the existing PointNet++network is unable to process large-capacity point cloud data,an al-gorithm based on improved PointNet++is proposed to realize the intelligent recognition of components for large-capacity hull segment convergence surface point cloud data.[Methods]Based on the hypervoxel growth theory,the hull segment closure point cloud data is segmented and simplified,and a hull segment clos-ure point cloud data set is constructed and used to train a PointNet++network improved by deep learning the-ory.[Results]The convergence results of the network model on the training and testing sets of hull seg-ment closure surface point cloud data tend to be stable,achieving an accuracy rate of 90.012%on the testing set.[Conclusions]The proposed method has good recognition ability and can achieve the intelligent recog-nition of hull segment closure surface components.
关键词
船舶建造/人工智能/船体分段合拢面/点云数据/超体素生长/PointNet++/智能识别Key words
shipbuilding/artificial intelligence/hull segment closure surface/point cloud data/hyper-voxel growth/PointNet++/intelligent identification引用本文复制引用
出版年
2024
国家科技期刊平台
NSTL
万方数据