YOLOv8_CB: An Improved YOLOv8 Model with CBAM and BiFPN for Pipeline Girth Weld Defect Detection

Girth weld defects, such as cracks, lack of penetration, lack of fusion, and edge nibbling, are among the primary causes of pipeline failure accidents. Traditional object detection methods face challenges in welding robot applications due to varying lighting conditions, complex backgrounds, and small target sizes. To address these issues, this study proposes an improved YOLOv8-based object detection method for welding robots in pipeline girth welding. The proposed method introduces the Convolutional Block Attention Module (CBAM) into the backbone network of YOLOv8 to enhance target features from both channel and spatial dimensions. Furthermore, a Bidirectional Feature Pyramid Network (BiFPN) structure is incorporated into the neck network to alleviate the problem of insufficient target feature information. Experimental results on the dataset demonstrate that the improved YOLOv8_CB model achieves a 3.07% increase in mean Average Precision () compared to the original algorithm. The proposed method provides a new perspective for accurate detection of pipeline girth weld defects and performance improvement of object detection in welding robot applications.