一种基于支持向量机的带钢表面缺陷识别方法

摘要

This paper proposed a classification and recognition method of cold steel strip based on SVM according to its good performance in small data sets and high dimension feature spaces. To classify several common defects of the steel strip, we used DAGSVM, and optimized the parameters and the classifiers by the cross-validation method. The results were compared with neural network algorithm .The results indicate that the algorithm of SVM applied to defect detection is better than the BP network in avoiding over-fitting and have better generalization ability. It has a good effect in surface defects recognition of cold steel strip.%鉴于支持向量机（SVM）在小样本、高维模式分类中具有的优良分类性能，可以基于多分类支持向量机来检测带铜表面的缺陷。本文构造了一类有向无环图支持向量机（DAGSVM），利用交叉验证进行了参数和模型的选取，对冷轧带钢中几种现场易出现的缺陷进行分类，并与BP神经网络进行比较分析。实验结果表明，这类基于SVM的算法识别效率较高，较好地解决了小样本学习问题，避免了BP神经网络出现的过学习、收敛速度慢、泛化能力弱等缺点，可有效地应用于带铜表面缺陷检测。