A Deep Learning Reconstruction Method for Fast Assembly Line Computed Tomography

The demand for mass production is increasing in the industrial field, which requires industrial non-destructive testing technologies to make a breakthrough for improving the efficiency of detection. However, the widely used industrial non-destructive testing tool, i.e. X-ray computed tomography (CT) techniques, usually adopt the one-by-one scanning mode and require a large number of projections to accurately reconstruct CT images, which greatly reduces the efficiency of CT and limits its application in industrial mass assembly line production. In this paper, we propose a deep learning reconstruction method for fast assembly line CT. It involves the fast assembly line CT scanning architecture and its corresponding deep learning reconstruction framework working in dual domains i.e. sinogram domain and image domain, which allows multi objects to be synchronously sparse-view scanned and obtain good-quality reconstruction images. The numerical simulation verifies that our method shows better performance in taking into account both the quality and efficiency of CT detection compared with other methods. This work will promote the applications of CT in industrial mass production.