曲面重构算法的研究  

Study on the Surface Reconstruction Algorithms

作　　者：熊邦书[1] 冯燕[1] 何明一[1] 吴铮[1] 惠增宏[1] 

机构地区：[1]西北工业大学电子工程系,西安710072

出　　处：《机械科学与技术》2003年第3期375-377,392,共4页Mechanical Science and Technology for Aerospace Engineering

基　　金：国家自然科学基金项目 ( 60 172 0 40 )资助

摘　　要：提出了一种基于多层感知器网络的曲面重构算法 (ML PSR) ,建立了用于曲面重构的多层感知器网络模型。本文对基于二维拉格朗日局部插值的曲面重构算法 (2 DL L SR)与 ML PSR进行对比。为了更好地评估这两种算法的有效性 ,选择可用数学表达式表达的曲面作为重构对象。仿真实验结果表明 ,这两种算法在曲面重构上各有优点。当 2 DL L SR算法的阶次高于等于曲面本身的阶次时 ,曲面重构的效果很好。ML PSR算法在曲面重构的速度和精度上都比较好 ,尤其在曲面受到破坏或不完全时特别有效。The paper presents a surface reconstruction algorithm based on multi layered perception(MLPSR), and constructs a neural network model of surface reconstruction. We first compared 2DLLSR algorithm based on local 2D Lagrange surface interpolation with MLPSR. To evaluate the effectiveness of the algorithms, a mathematically known surface was used to generate a number of samples to train the neural network and to be severed as known points of 2DLLSR. The experiment results show that the precision of 2DLLSR is high when the order of 2DLLSR is higher than that of the reconstructed surface. However, there is a tradeoff between the speed of reconstruction and the precision. The MLPSR algorithm solves this contradiction better, and it is particularly useful for the situation where the surfaces are partial or the surface models are not available.

关 键 词：曲面重构 MLP网络 局部二维拉格朗日曲面重构算法 逆向工程 

分 类 号：O174.21[理学—数学] TP183[理学—基础数学]