基于深度学习的船体三维模型自动生成方法  

3D Ship Model Generation Algorithm Based on Deep Learning

作　　者：王晓琦 赵旸 张键 王硕[2] WANG Xiaoqi;ZHAO Yang;ZHANG Jian;WANG Shuo(CSSC Systems Engineering Research Institute,Beijing 100094,China;School of Electrical and Information Engineering,Zhengzhou University,Zhengzhou 450001,Henan,China)

机构地区：[1]中国船舶集团有限公司系统工程研究院,北京100094 [2]郑州大学电气与信息工程学院,河南郑州450001

出　　处：《兵工学报》2022年第S02期115-119,共5页Acta Armamentarii

基　　金：中国船舶集团有限公司系统工程研究院科研项目(2021年)。

摘　　要：三维船体模型具有广泛的应用场景,但制作成本较高。提出一种基于深度学习的船体三维模型自动生成方法:通过卷积神经网络结构从一张二维平面图中提取角点;按照特定的规则将各角点组成线段,每一个线段表示一面墙体;将线段变换生成三维模型。制作500张二维船图用于模型训练,并制作30张二维船图用于测试。研究结果表明:新算法可以精确地检测输入图片的角点位置和种类,并生成清晰、美观、通用的三维模型,效果很好。The 3D ship model has a wide range of application scenarios,but its production cost is high.We propose an automatic generation method of 3D ship model based on deep learning:corner points are extracted from a 2D plan with the convolution neural network structure;each corner point is formed into a line segment according to specific rules,and each line segment represents a wall;the line segments are transformed to generate the 3D model;5002D ship diagrams are made for model training,and 302D ship diagrams for testing.The results show that the new algorithm can accurately detect the corner positions and types of the input image,and generate clear,beautiful and universal 3D models,and the test results demonstrate that the proposed method is effective.

关 键 词：船体三维模型 深度学习 船图 卷积神经网络 

分 类 号：U662.9[交通运输工程—船舶及航道工程]