基于PSO-SVM的带钢表面缺陷分类研究

Study of Strip Surface Defect Classification Based on PSO-SVM

带钢表面缺陷分类是一个多分类问题,采用传统分类器进行分类识别率较低.建立PSO优化SVM模型,对带钢表面分类问题进行了系统研究.模型采用交叉验证法,针对UCI数据库的带钢表面缺陷生产数据进行了实例分析,该方法可对1941组生产数据中的7种不同类型缺陷进行较为准确的分类.经过与传统SVM、BP神经网络等方法进行对比,PSO-SVM体现出了更高的准确性和泛化能力,对生产实际有一定的指导作用.

Strip surface defect classification is a multi-classification problem, traditional classifiers are faced with low accuracy on this problem. SVM model optimized by PSO is implemented for systematical research of strip surface defect classification problem. In the model,K-CV statistical analysis is used and the production data provided by UCI database is analyzed. The result shows that by using the PSO-SVM model, the whole 1941 sets of data with 7 different kinds of defects have been classified pretty accurately. Compared with traditional SVM, BP neural networks and some other methods,PSO-SVM shows a higher accuracy and better generalization ability, which plays a guiding role on actual production.