Abstract: Problems of signal pattern recognition for defects identification in ultrasonic inspection are analyzed. A new fusion decision-making method base on multiple features extraction is presented, which compounds with support vector machine theory and evidence theory. The principles of SVM method and the Dempster fusion are introduced. The concepts of DS framework are referred to defect identification, and the fusion decision-making rules are designed to resolve identification of multiple defects with features extracted from different ways. Four features extraction methods from different spatial domains of a signal are presented for fusion decision-making recognition. Experiments with both SVM method and SVM-DS method are carried out to identify the four typical defects in oil casing pipes. The result shows that the said typical defects can be identified effectively by SVM-DS method, and both recognition rate and generalization are better than SVM method.

Key words: Defect identification, Evidence theory, Fusion decision-making, Support vector machine