基于支持向量机（SVM）方法和模拟油田现场管道磁记忆检测数据分别建立了管道缺陷的分类识别和分级识别方法；采用该方法对油田现场5根在役油气管道的缺陷类型进行了识别，并对试验管道的穿孔腐蚀与未穿孔腐蚀两种腐蚀程度进行了识别。结果表明：以三类不同特征量的组合分别建立的SVM模型对缺陷类型的识别率分别是77.08%、89.58%和95.83%，其中使用时域、形态和频域特征量的SVM模型的识别率最高；腐蚀缺陷分级识别方法的识别率达到了90%。该方法可有效识别管道腐蚀缺陷和应力集中缺陷，以及腐蚀缺陷的腐蚀程度。

Based on support vector machine (SVM) method and magnetic memory detection data of pipeline in simulated oilfield scene, identification methods for pipeline defect classification and grading were established. The defect types of 5 oil and gas pipelines in service were identified and two degrees of corrosion, perforated corrosion and non-perforated corrosion, of the test pipelines were identified by the methods. The results showed that the identification rates of defect classification were 77.08%, 89.58% and 95.83% respectively for the SVM models established with different combination of three types of characteristic quantities, in which the identification rate of the model based on characteristic quantities of time domain, form and frequency domain was the highest. In addition, the classification rate of defect grading was 90%. The methods could effectively be used to identify corrosion defects and stress concentration defects, and to grade corrosion defects.