Defects in girth welds are the main factor causing pipeline failure. Among numerous non-destructive testing technologies, Digital Radiography non-destructive testing technology has become the main non-destructive testing technology due to its high recognition rate and simple equipment. However, the high degree of manual intervention and low recognition rate have become the main factors restricting its development. Therefore, the artificial intelligence recognition of Digital Radiography detection is researched in this article. Firstly, multiple denoising processes are applied to actual Digital Radiography detection images, and the mean filter with a kernel size of 3*3 is obtained to achieve the best denoising effect. Subsequently, various enhancement processes are applied to the denoised image. A linear transformation with a = 0.7 and a Gamma transformation with γ = 2 has the best Enhancement effect. The Global Attention Mechanism and Bidirectional Feature Pyramid Network are added to the YOLOv8 algorithm to adapt to defect recognition. The improved yolov8 algorithm achieved a recognition rate of 83.1%, which is 10.7% higher than the yolov8 algorithm. Therefore, this method fills the gap in intelligent image recognition of Digital Radiography in pipeline girth welds.