改进组合分类器的冷轧带钢表面缺陷识别研究
Study on Surface Defect Recognition of Cold Rolled Steel Strip by Improving Combinationclassifier

针对表面缺陷种类多样、形态复杂的冷轧带钢，若采用单一分类器识别分类，会存在对个别缺陷不敏感、识别率低的情况，且会导致分类器处理特征数据规模过大，系统的鲁棒性和稳定性很难得到保证。为此提出基于改进组合分类器的冷轧带钢表面缺陷识别方法，将优化BP神经网络、概率神经网络以及改进的支持向量机进行组合，利用分类信息的互补性进行综合分类，从而构建了较优的分类系统。实验结果表明：改进组合分类器弥补了单个分类器网络训练的不足；针对每一类缺陷识别时准确率都较高，能增加整体分类器的泛化能力，整体识别正确率可达95%以上，且识别高效、稳定，具有实用价值。
For the cold-rolled strip that has various types of surface defects and complex form, there will be insensitive individual defects and low recognition rate if a single classification is used to identify classification. Therefore it probably results in data classification processing feature size is too large, the robustness and stability of the system is difficult to ensure. The method based on the the improved combination classifier is put forward. It will optimize BP neural network and probabilistic neural network, combining the improved support vector machine. Moreover, it will use complementary classification information to classify. Optimum classification system can be constituted. The experimental results show that the combination of the improved classifier makes up for the lack of a single classifier Network training. For each type of defect recognition, the accuracy is high. It can increase the overall classification generalization ability. The recognition accuracy rate is more than 95%. In short, there is efficient recognition and practical value