基于极端学习机的胶囊缺陷检测  

Inspection for defected capsules based on extreme learning machine

作　　者：赖大虎[1] 黄宴委[1] 

机构地区：[1]福州大学电气工程与自动化学院,福建福州350108

出　　处：《福州大学学报（自然科学版）》2012年第4期489-494,共6页Journal of Fuzhou University(Natural Science Edition)

基　　金：教育部博士点新教师基金资助项目(No.20113514120007);福建省自然科学基金资助项目(2010J05132);福建省教育厅科研资助项目(JA10034)

摘　　要：针对胶囊常见的凹陷缺陷进行不变矩分析,提取胶囊的偏心率和圆形度为特征变量,提出一种基于极端学习机的胶囊缺陷检测与识别的方法,对胶囊进行检测分类.仿真结果表明基于极端学习机的胶囊分类算法能很好地区分出具有缺陷的胶囊,分辨正确率接近100%,运算速度比BP神经网络更快,训练过程稳定,也说明所提取的特征能很好区分有凹陷与无凹陷胶囊.In this paper, considering invariant moment for common dent capsules and a method of inspection based on extreme learning machine （ELM） is presented to detect for defected capsules, in which eccentricity and circularity of the capsule are extracted as the features. Simulations indicate the defected capsules can be detected with an approximate accuracy of 100%. The operating speed of ELM is a lot better than that of BP neural network and the detection performance is stable. The results also prove our features are effective to classify the capsules.

关 键 词：胶囊 特征变量 极端学习机 缺陷 检测 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]