Abstract: The electrofusion welding status of the polyethylene (PE) gas pipeline can be obtained from the ultrasonic pictures taken by the phased array system. However, whether there are welding defects or not is judged by professionals manually identifying the information related to defects such as feature line, resistance wire, and bottom echo line in each picture, so as to determine the defect category and grade. This method is time-consuming, laborious, and prone to missed detections and false detections. Aiming at the identification of electrofusion welding defects in PE pipelines, this paper proposes an automatic identification method of welding defects based on image processing technology, which can judge the categories and grade of defects in ultrasound images at the same time. The method consists of four steps: (1) expanding the number of existing pictures through data enhancement technology to build a data set; (2) training the image semantic segmentation model to segment the image semantically; (3) using mathematical morphology to remove the noise of the segmentation result, and obtaining defect-related information via the connected domain analysis; (4) identifying the defect category and grade according to the welding standard and defect-related information. Finally, the experimental results show that the proposed defect recognition method can meet the requirements of industrial applications in terms of accuracy, recall and running time.