Abstract: Using deep learning network of Faster R-CNN and YOLO, we performed defect recognition for radiographic image of China-Russia Eastern Gas pipeline. In this study, the defect sample database of automatic welding girth welds was established, and the research of defect recognition for radiographic image was completed, which preliminarily realized the intelligent identification of hazardous defects such as lack-of-fusion and cracks. Under the construction target of China-Russia Eastern Gas Pipeline, the development and application of new artificial intelligence technologies such as radiographic image recognition can help to unearth the value of pipeline big data and improve the level of pipeline intelligent operation and management.