A fast weld region segmentation method with noise removal

Weld region segmentation (WRS) is a key step in the recognition of defects from radiographic testing (RT) to ensure manufacturing and service quality of large-scale equipment. Traditional methods for WRS are time consuming and susceptible to image noise, particularly defect edges in the weld region. To address these issues, a fast WRS method is proposed in this paper to effectively remove the noise in RT images. First, a new image segmentation method based on beetle antennae search and clustering analysis is applied to segment an RT image and rapidly obtain a binary image. Second, a weld region boundary segmentation method is presented to eliminate noise based on morphological filters, remove defect points using object deletion, and obtain a continuous and smooth boundary of the weld region. A collection of radiographic weld images from a steam turbine manufacturing enterprise are then used to validate the proposed method. The results show that the proposed method has advantages in terms of reduced computation time and complexity compared to traditional methods. Furthermore, the WRS using the proposed method exhibited little image noise and few local defect points, accurately reflecting the boundary of the weld region both with and without defects.