船体曲面特征的计算机视觉认知与生成机制研究  

Ship hull feature recognition and creation functionality investigation based on computer vision technology

作　　者：杜林 李广年 郑彭军[1] DU Lin;LI Guang-nian;ZHENG Peng-jun(Ningbo University,Ningbo 315000,China)

机构地区：[1]宁波大学,浙江宁波315000

出　　处：《船舶力学》2022年第7期949-961,共13页Journal of Ship Mechanics

基　　金：高等学校学科创新引智计划(D21013);装备预先研究领域基金(80907010201)。

摘　　要：船体曲面的型值数据与图片数据都是序列相关型数据,即数据排列形式与所表达信息紧密相关的数据类型,因此用于图片分类和生成的计算机视觉技术,理论上也适用于船体曲面特征的认知与生成。本文基于此共性假设,对基于计算机视觉的船体曲面特征认知与生成机制开展系列研究。通过卷积神经网络搭建多元分类器用于识别不同区域的船体曲面形状,在多种曲面分割方案下均得到较为理想的分类精度;基于生成式对抗网络模型,设计一体式和分区式船体曲面生成方法,并根据这两种方法分别尝试进行曲面生成和重构,证明两种方法都能依据其自身特点部分实现船体曲面生成功能,为后续的研究提供参考。本文研究结果可为船体曲面快速建模与变形方法方面的研究提供技术和理论支持。As pictures, the ship-hull offset data are sequential in space, and the hull geometry information is strictly associated with the sequence of values. Due to this point, the ship hull’s cognitive and creative functionality was investigated based on computer vision technology. Firstly, a multi-classifier was established with the convolutional neural networks to distinguish the different ship hull areas with ideal classification accuracy. Furthermore, two ship-hull surface creation methods, the all-in-once method and the separative method, were developed with generative adversarial networks. The result of training and testing indicates that the two methods can generate ship hull surfaces with pros and cons, providing theoretical and practical support for investigation of ship hull modeling and transformation based on computer vision.

关 键 词：船体曲面建模与变形方法 卷积神经网络模型 生成式对抗网络模型 计算机视觉技术 

分 类 号：U662[交通运输工程—船舶及航道工程]