基于局部相似度特征的管道超声内检测异常信号识别方法

Ultrasound anomaly signal identification method based on local similarity feature

超声异常信号识别对于降低缺陷误检,提升缺陷检测、反演精度具有至关重要的作用。然而由于超声异常信号模式复杂多样,样本标注无法穷尽其模式,导致当前异常信号识别方法的精度较低,为进一步提升超声异常信号的识别精度,提出了一种基于局部相似度特征的超声异常信号识别方法。首先,设计了一种基于动态时间规整的局部相似度特征提取方法,该方法在不使用超声异常信号样本的情况下依然能够提取到有效可分的特征。然后,在局部相似度特征向量的基础上,利用单类支持向量描述算法进行异常识别。实验结果表明,该方法的受试者工作曲线下面积为0.995,相较于传统无监督异常识别算法提升0.18~0.61,在实际工程中具有较强的应用价值。

The recognition of ultrasound abnormal signals plays a crucial role in reducing defect detections,improving defect detection and inversion accuracy.However,due to the complex and diverse patterns of ultrasound abnormal signals,sample labeling cannot exhaust their patterns,resulting in low accuracy of current abnormal signal recognition methods.To further improve the recognition accuracy of ultrasound abnormal signals,a local similarity based ultrasound abnormal signal recognition method is proposed.Firstly,a local similarity feature extraction method based on dynamic time warping was designed,which can still extract effective and separable features without using ultrasound abnormal signal samples.Then,based on the local similarity feature vectors,a single class support vector description algorithm is used for anomaly recognition.The experimental results show that the area under the working curve of the subjects using this method is 0.995,which is 0.18-0.61 higher than traditional unsupervised anomaly recognition algorithms.It has strong application value in practical engineering.