摘要
针对带钢表面缺陷识别问题,提出一种基于动态演化复杂网络特性的特征描述方法,这些特征同时具有位移、旋转不变性、大小不变性、较强的抗干扰能力和鲁棒性,为缺陷识别提供良好的分类特征;为了提高分类器的效率,应用主成分分析法(Principal component analysis,PCA)对复杂网络特征向量进行特征降维处理;采用最优有向无环图支持向量机(Directed acyclic graph support vector machine,DAG-SVM)算法进行缺陷分类.结果表明该方法识别率高而且识别速度快.
A feature extraction method based on the characteristics of dynamic evolution complex networks is proposed for the strip steel surface defect recognition. The extracted features possess displacement, rotation and size invariability, strong antiinterference ability and robustness, therefore they are good classification features for steel surface defect recognition. In order to improve the efficiency of classification, the principal component analysis （PCA） is adopted to reduce the dimension of the feature vector. The directed acyclic graph support vector machine （DAG-SVM） algorithm is used for the defect classification. The experimental results show that this method is of high recognition rate and fast recognition speed.
出处
《自动化学报》
EI
CSCD
北大核心
2011年第11期1407-1412,共6页
Acta Automatica Sinica
基金
国家自然科学基金(60804040)
霍英东教育基金会(111065)资助~~
关键词
缺陷识别
复杂网络特征
主成分分析法
有向无环图支持向量机
Defect recognition
complex network characteristics
principal component analysis （PCA）
directed acyclic graph support vector machine （DAG-SVM）
