摘要
船体分段合拢面的精度检测是分段总组合拢过程中的重要环节。在船体分段合拢面的精度检测方面,三维扫描仪相对全站仪有着巨大优势,但三维扫描仪在扫描过程中会记录很多与合拢面无关的点。文章对三维扫描仪扫描出的点云数据进行合拢面的智能识别;采用深度学习理论对PointNet++点云网络进行改进,使用CAD模型导出的点云数据构建有标注的船体分段点云数据集,进而使用Adam优化算法对网络进行优化训练。最终,网络模型对分段合拢面的识别在验证集上获得精确率73%、召回率90%的效果。
The accuracy detection of block erection surface is an important part of assembling and erection process.In terms of the accuracy detection of block erection surface,the 3D scanner has a huge advantage over the total station.However,the 3D scanner records many points that are not related to block erection surface during the scanning process.Therefore,the block erection surface is intelligently recognized by point cloud data scanned by 3D scanner.Appropriate improvements have been made to the PointNet++network according to the deep learning theory.The point cloud data derived from the CAD model is used to construct the labeled point cloud data set,and then the Adam algorithm is used to optimize the network.Finally,the network’s recognition of block erection surface achieves 73%precision and 90%recall on validation data set.
作者
陈尚伟
汪骥
刘玉君
张学晨
CHEN Shangwei;WANG Ji;LIU Yujun;ZHANG Xuechen(School of Naval Architecture and Ocean Engineering,Dalian University of Technology,Liaoning Dalian 116024,China)
出处
《船舶工程》
CSCD
北大核心
2019年第12期138-141,共4页
Ship Engineering
基金
辽宁省高等学校创新团队项目(LT2014002).
作者简介
陈尚伟(1994—),男,研究生,研究方向:基于三维扫描点云的船体合拢面识别与定位。
