基于Grouplet熵与关联向量机的断口图像识别方法研究  

A Recognition Method of Fracture Images of Aerial Material Based on Grouplet Entropy-RVM

作　　者：孙熠[1] 李志农[1] 闫敬文[2] 

机构地区：[1]无损检测技术教育部重点实验室(南昌航空大学),南昌330063 [2]广东省数字信号与图像处理技术重点实验室(汕头大学),广东汕头515063

出　　处：《失效分析与预防》2015年第1期1-5,共5页Failure Analysis and Prevention

基　　金：国家自然科学基金(51261024;51075372)

摘　　要：Grouplet变换是通过Haar变换实现的一种二维图像多尺度分析技术,拥有根据图像的纹理结构自适应改变基的能力,从而具有较好的稀疏性。与小波变换相比,Grouplet变换在针对纹理复杂的金属断口图像的识别方面具有更优越的性能;将Grouplet变换与关联向量机结合,采用Grouplet熵作为特征,关联向量机作为识别器,提出了一种新的基于Grouplet熵-RVM的航空构件断口图像识别方法。试验表明:该方法结合了Grouplet变换以及关联向量机的优势,在针对222张断口图像的训练与识别中,识别率达到了85.58%,相比Grouplet熵-SVM方法识别速率提高了5倍。Grouplet transform is a multi-scale analysis technology of two-dimensional image through Haar transform, which has the ability to adaptively change the basis when the image texture has changed. Compared with the wavelet transform, Grouplet transform has better ability in sparse representation and it has more superior performance in to deal with the complicated metal fracture of texture image recognition. Proposed a new recognition method based on Grouplet entropy and relevance vector machine. In the proposed method, Grouplet entropy is used as the image’ s feature, RVM as a classifier. The experimental results show that this method combined the advantages of Grouplet and RVM. The proposed method get 85. 58% of correct recognition rate for 222 metal fracture images. Compared with Grouplet entropy-SVM, the recognition speed of Grouplet entropy-RVM is about five times than Grouplet entropy-SVM.

关 键 词：Grouplet变换 关联向量机 特征提取 金属断口 图像识别 

分 类 号：TG115.2[金属学及工艺—物理冶金]