摘要
基于图像信息的缺陷识别技术是带钢表面缺陷检测系统中的关键技术之一。通过采用单一的分类技术或者一步到位的创建分类器,对复杂带钢表面缺陷图像进行识别很难达到理想的效果。提出了用Boosting算法结合SLIQ决策树建立组合分类器来识别带钢表面缺陷的方法。Boosting算法通过适应性权重技术和带权重的投票方法,建立并组合多个功能互补的分类器,组合分类器通过优势互补的方法有效地提高单个分类器的性能;而SLIQ决策树算法的数据预排序和广度优先技术对大规模数据分类具有速度优势,适合于作为单个分类器的弱学习算法。对实际带钢表面缺陷数据集进行测试,Boosting优化SLIQ决策树的组合分类器对缺陷识别的准确率达到了90%以上。
Defect recognition based on image information is one of the key technologies in strip steel surface automatic inspection system.Applying a single type of classification technology or building classifier in one step,image classification of complex strip steel defects is very difficult to achieve the desire result.In this paper,a combination classifier based on Boosting algorithm and SLIQ decision tree was presented.With adaptive weight updating and weighted voting,Boosting algorithm could build and combine multiple functional complementary classifiers,and improve recognition performance effectively compared with a single classifier.Through data pre-sorting and breadth-first growing methods,SLIQ decision tree was fit for large data processing and could be used in weak learning algorithm of each single classifier.Tests on real strip steel surface defect datasets were carried out.The result shows that combination classifiers of Boosting optimized SLIQ decision tree can improve the accuracy of defects recognition up to 90%.
出处
《红外与激光工程》
EI
CSCD
北大核心
2010年第5期954-958,共5页
Infrared and Laser Engineering
基金
国家自然科学基金资助项目(60736010)
