基于局部形状特征和Bag-of-Feature模型的磨粒图像形状特征提取  

Wear Particle Image Shape Feature Extraction Based on Local Shape Feature and Bag-of-Features Model

作　　者：鲁华杰 张伟 刘涛 LU Huajie;ZHANG Wei;LIU Tao(College of Coastal Defense Force,Naval Aviation University,Yantai 264001;Vocational Education Center,Naval Aviation University,Yantai 264001)

机构地区：[1]海军航空大学岸防兵学院,烟台264001 [2]海军航空大学职业教育中心,烟台264001

出　　处：《舰船电子工程》2021年第4期27-30,155,共5页Ship Electronic Engineering

摘　　要：以磨粒图像为研究对象,提出了基于局部形状特征和Bag-of-Features模型的磨粒形状特征提取方法。首先构建磨粒区域的骨架,根据骨架端点和分支得到磨粒的轮廓基元和区域基元,不同数量的相邻的轮廓和区域基元组合构成局部轮廓和局部区域,提取局部轮廓和区域的形状特征,两者融合得到磨粒的局部形状特征集合。然后根据Bag-of-Features模型的思想,以训练集所有磨粒样本的局部形状特征集合为基础,构建视觉词典,经过特征编码、特征汇集和归一化,得到磨粒形状特征的编码向量表示。最后根据形状特征的编码向量,采用多级支持向量机的方法对磨粒类型进行识别。实验结果表明,基于提出的磨粒形状特征方法能够有效、准确地识别磨粒类型。Taking wear particle image as research object,wear particle shape feature extraction method based on local shape feature and Bag-of-Features model is proposed. Firstly,the skeleton of wear particle is constructed,and according to the endpoints and branches of skeleton,the contour and region primitives of wear particle are acquired. The local contour and local region areformed by combining adjacent contour and region primitives separately,and the local shape feature set is got by fusing local contour and region shape features. Secondly,according to the Bag-of-Features model,the visual dictionary is constructed based on wear particle local shape feature sets of training set. After feature encoding,feature pooling and normalization,encoding vector representation of wear particle shape feature is acquired. At last,according to the encoding vector of wear particle shape feature,multi-class support vector machine is used to recognize the wear particle type. The experiment results show that wear particle type can be recognized effectively and accurately based on the proposed wear particle shape feature.

关 键 词：磨粒图像 形状特征 骨架 Bag-of-Feature模型 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置]