摘要
分析了超声检测信号识别中存在的问题。研究了将支持向量机和贝叶斯推理相结合的多特征融合识别算法。阐述了支持向量机解决分类问题的原理以及贝叶斯推理原理。设计了基于最大后验概率准则的多缺陷类型多特征SVM-Bayes融合识别方法。介绍了四种不同的特征提取方法。分别将单特征SVM方法和SVM-Bayes融合方法应用于石油套管缺陷检测信号的识别。对比试验表明:SVM-Bayes融合识别方法能有效识别上述缺陷信号,其在识别率和泛化性方面都比单特征的SVM识别方法有优势。
Problems in ultrasonic signal recognition were analyzed. A new fusion recognition method base on multi- features extraction was studied, it was combined with support vector machine （SVM） theory and Bayes reasoning. The principles of SVM method and Bayes reasoning were introduced. The fusion recognition method based on maximum a posteriori （MAP） was designed to identify signals of different defects with features extracted in different ways. Four feature extraction methods were presented for fusion recognition. Experiments with both SVM method and SVM-Bayes one were carried out to identify defect signals of oil casing pipe. The result showed that defects can be identified effectively with SVM-Bayes method, and its recognition rate and generality are better than those of a single feature SVM method.
出处
《振动与冲击》
EI
CSCD
北大核心
2011年第12期265-269,共5页
Journal of Vibration and Shock
关键词
支持向量机
贝叶斯推理
融合识别
support vector machine （SVM）
Bayes reasoning
fusion recognition
