Abstract: As variety of surface defect types of cold rolled strips，it is very impendent to find a group of optimized feature set among original feature set extracted from surface defects，which can express essence characters of defects more effectively and improve defect recognition rate.For this problem，genetic algorithm is researched here for the application to defect feature selection on the basis of fully research of theory of information entropy，and average net classification information is used as a fitness function of genetic algorithm to make up for the shortage of using mutual information entropy as fitness function of genetic algorithm.Experiments show that using optimized feature by genetic algorithm to classify surface defects of cold rolled strips，a higher accurate recognition rate can be achieved.

Key words: genetic algorithm, feature selection, information entropy, cold rolled strip, surface defects

摘要： 由于冷轧带钢表面缺陷的类型多种多样，在所提取的特征集中，需要寻找出一组较优的特征集，使之可以更有效地表达缺陷的本质特征，从而提高缺陷识别的准确率。针对该问题，研究了遗传算法在缺陷特征选择中的应用，并在充分研究信息熵理论的基础上，以平均净分类信息为遗传算法的适应度函数，以弥补互信息熵作为适应度函数所导致的不足。实验表明，利用遗传算法得到的特征集，对现场的冷轧带钢表面缺陷进行分类时，能得到更高的分类准确率。

关键词: 遗传算法, 特征选择, 信息熵, 冷轧带钢, 表面缺陷