• 中国计算机学会会刊
  • 中国科技核心期刊
  • 中文核心期刊

计算机工程与科学

• 论文 • 上一篇    下一篇

基于支持向量机的枪弹外观缺陷识别与分类

王鹏,郭朝勇,刘红宁   

  1. (军械工程学院车辆与电气工程系,河北 石家庄 050000)
  • 收稿日期:2015-07-31 修回日期:2015-09-29 出版日期:2016-09-25 发布日期:2016-09-25

Bullet surface defect recognition and classification  based on support vector machine      

WANG Peng,GUO Chao-yong,LIU Hong-ning   

  1. (Department of Vehicle and Electrical Engineering,Ordnance Engineering College,Shijiazhuang 050000,China)
  • Received:2015-07-31 Revised:2015-09-29 Online:2016-09-25 Published:2016-09-25

摘要:

为解决枪弹外观缺陷自动分类问题,提出了一种基于支持向量机的枪弹外观缺陷自动识别与分类模型。首先针对枪弹表面缺陷的图像特点,从几何、灰度、纹理三方面进行了特征提取,在此基础上建立了基于支持向量机的枪弹外观缺陷分类模型,并对特征参数进行了优选;研究了支持向量机中惩罚系数和核函数参数对分类器性能的影响;通过实验与基于BP神经网络的枪弹外观缺陷分类器进行了比较,结果表明,在小样本下,基于支持向量机的枪弹外观缺陷分类器性能更好。

关键词: 支持向量机, 枪弹外观缺陷, 特征参数, 识别与分类

Abstract:

In order to solve the problem of bullet surface defect automatic classification, we propose an automatic recognition method and a classification model of bullet surface defects based on the support vector machine (SVM). Firstly we extract the characteristic parameters from the three aspects of geometry, gray scale, texture according to the characteristics of the bullet surface defect image. Then we establish the classification model of bullet surface defects based on the SVM and the characteristic parameters are optimized. We also analyzed the influence of the penalty coefficients and kernel function parameters on the performance of the classifier. Experimental results show that the proposal based on the SVM outperforms the BP neural network classifier in terms of bullet surface defect classification under small samples.

Key words: support vector machine, bullet surface defect, characteristic parameters, recognition and classification