基于GA-SVM的带钢表面缺陷模式识别

摘要

带钢的表面缺陷模式识别是一个生产当中遇到的多分类问题。传统分类器面对多分类问题往往出现泛化性能差，过度学习，识别率低等问题。本文采用遗传算法（GA）对支持向量机（SVM）分类器进行参数优化，并将其应用于带钢表面缺陷分类的模式分类。实验数据来源于UCI标准数据集，采用10折交叉验证法进行分类仿真。通过与传统SVM分类器以及BP神经网络等方法进行对比，GA-SVM模型的分类准确率更高，对带钢表面缺陷多分类问题有一定指导作用。%Strip surface defect classification is a multi-classification problem encountered in the production. Traditional classifiers are often faced with the problem like poor generalization performance, excessive learning and low accuracy. In this paper, SVM model optimized by GA was implemented for systematical research of strip surface defect classification problem. In the paper, UCI database was implanted as experimental data, and 10-fold cross-validation was used for simulation. The result showed that GA-SVM model worked better on accuracy than traditional SVM, BP neural networks and some other methods, which had a guiding role on strip surface defect classification.