Abstract: Due to the small size of small hardware target,it is difficult to obtain the vertical/horizontal deformation maps of target details, which easily leads to misidentification. To address this issue, this study proposed an automatic defect identification method for live working hard hardware based on G-SAD2Dec and X-ray flaw detection technology. Based on the attenuation degree of X-rays passing through materials, high-resolution internal structural images of hard hardware were obtained through X-ray flaw detection technology. G-SAD2Det captures the subtle vertical/horizontal deformation characteristics of small hardware targets in images by introducing a global attention mechanism and multi-scale vertical/horizontal deformation fusion, learns the target box offset through an independent deep learning method to optimize the target box position, and realizes automatic identification of hardware defects. The experimental results show that the method identifies bolt loosening defects in the horizontal range of 0-4 cm and vertical range of 4-6 cm,nut missing defects in the horizontal range of 2-8 cm and vertical range of 7-9 cm, pin missing defects in the horizontal range of 5-8 cm and vertical range of 5-7 cm, pin loosening defects in the horizontal range of 3-8 cm and vertical range of 5-7 cm, and has accurate identification effect.