项目名称: 基于Grouplet变换的航空构件断口图像识别新方法研究
项目编号: No.51261024
项目类型: 地区科学基金项目
立项/批准年度: 2013
项目学科: 一般工业技术
项目作者: 李志农
作者单位: 南昌航空大学
项目金额: 50万元
中文摘要: Grouplet变换是一种非常新颖的基于图像几何流最佳稀疏表示的正交变换,其性能大大优于传统小波和其他方向性小波。在此,将Grouplet变换引入到金属断口图像识别中,提出了基于Grouplet变换的航空构件断口图像识别新方法, 该方法的特色在于:1)增强了对金属断口图像的自适应处理能力,可以最大限度地充分利用金属断口图像的几何特征。2) 提出的Grouplet峭度和Grouplet盒维数的概念,具有明显的物理意义,大大增强了对金属断口图像的特征提取的能力。3)将稀化核主分量分析引入到金属断口图像识别中,把金属断口图像非线性识别转化为线性识别,在不降低识别效果的前提下大大简化了金属断口图像识别的复杂性。4) 提出的基于Grouplet变换的关联向量机识别方法具有较好的泛化能力和在核数量不增加的情况下可以大幅度提高金属断口的识别率。本项目所取得的研究成果必将对失效分析研究起着很大的推动作用。
中文关键词: Grouplet变换;金属断口;特征提取;模式识别;失效分析
英文摘要: Grouplet transform is a very novelty orthogonal transform based on the optinum sparse expression of image's geometry stream, its performance is greatly superior to the traditinal wavelet and the other directional wavelet. Here, Grouplet transform is introduced into the recognition of metal fracture images, a new recognition method of aviation components fracture images is proposed. The features of the proposed method are as follows: 1) The self-adaptive processing ability of metal fracture is enhanced, this method can make maximum use the geometrical characteristics of metal fracture. 2) The proposed concept of Grouplet kurtosis and Grouplet bosx dimension have the obvious physical meaning, the featrure extraction of metal fracture images is greatly enhanced. 3) The sparse Kernel principal component analysis is introduced into the recognition method of metal fracture images, the nonlinear recognition of metal fracture images is changed into the linear recognition,thus greatly simplify the complexity of the metal fracture image recognition under the recognition effect is not reduced. 4) the proposed relevance vector machine recognition method based on Grouplet transform has the better generalization ability and greatly improves the recognition rate of metal fracture under the kernel number does not increase. The
英文关键词: Grouplet transform;Metal fracture;Feature extraction;Pattern recognition;Failure analysis