论文部分内容阅读
针对射线检测焊缝图像中缺陷识别正确率低的问题,提出一种选择性集成学习的焊接缺陷识别算法.算法中的个体学习器由稳定分类器和非稳定分类器组成,使用SVM-RFE算法移除集成学习器中的冗余个体学习器,保留子学习器预测输出加权作为集成学习器的输出,有效地增强了个体之间的差异性,进而提高了集成的泛化性能.结果表明:该算法充分利用更多的缺陷特征和样本数据集信息,继承了强集成学习的优点,有效地提高分类正确率.使用一对多的方法把二分类选择性集成学习器推广到多分类问题中,所提出的算法在训练精度为92.4%时;焊缝缺陷识别率提高到85.5%.
In order to solve the problem of low accuracy of defect detection in ray-detected weld images, a selective welding defect recognition algorithm based on integrated learning is proposed.The individual learners in the algorithm are composed of stable classifiers and unsteady classifiers, SVM-RFE The redundant individual learners in the integrated learner are removed and the predictive output weight of the learner is reserved as the output of the integrated learner, which effectively enhances the individual differences and enhances the generalization performance of the integrated learners.The results show that: The algorithm takes full advantage of more defect features and sample dataset information, inherits the advantages of strong integrated learning and effectively improves the accuracy of classification.Using the one-to-many method to generalize dichotomous selective integrated learners into multi-class problems The proposed algorithm has a training accuracy of 92.4% and the recognition rate of weld defects increases to 85.5%.