Abstract: With the combination of radiographic evaluation system and artificial intelligence technique based on big data, the intelligent filter of invalid films and duplicate films and the intelligent identification and evaluation of weld defects were realized. It not only effectively improves the deficiency of the method of traditional inspection, but also enhances the quality of inspection and management. In this paper, the gordian methods of image segmentation model in the field of deep learning and classification and identification of weld defects and extraction of comprehensive information of weld and the modules of core algorithm were described. And the feasibility and stability of the radiographic evaluation system were verified by practical engineering projects, which could be a reference for the subsequent large-scale application.