基于混合染色体的带钢缺陷图像分类方法研究

Strip steel surface defects classification based on hybrid chromosome

为了保证带钢缺陷分类的实时性和准确性,提出了一种基于混合染色体的带钢缺陷图像分类方法。该方法不仅优化了支持向量机SVM中核函数参数、惩罚因子,并且对核函数、输入的特征向量进行了选择。除此之外,该方法融合了遗传算法和SVM,用遗传算法优化影响SVM的核函数参数、惩罚因子、输入特征和核函数;同时,用SVM建立的分类模型的分类准确率限制遗传算法的进化方向,彼此制约和促进,最终确定最优分类模型。实验结果表明,基于混合染色体的带钢缺陷图像分类方法建立的分类模型能实时、准确地对带钢缺陷图像进行分类。

In order to guarantee the characteristics of real time and accuracy in strip steel defect clas- sification, we propose a defect classification model based on hybrid chromosome for the strip steel surface image. This method not only optimizes both the kernel function parameters and the penalty factors of the support vector machine （SVM） model, but also makes choice of feature vectors and kernel functions. Meanwhile, the genetic algorithm and the SVM model are fused and the final SVM classifier is established by using the decoding results of the optimal chromosome in the end. Experimental results show that our method is effective and efficient in classifying the defects in the strip steel surface image.