Fast reconstruction of CFRP X-ray images based on a neural network filtered backprojection approach

X-ray inspection of Carbon Fiber Reinforced Polymer (CFRP) is time consuming since many projections at different positions around the object are required to obtain reconstructions of the interior fiber composition of reasonable quality with classical reconstruction methods. In this work, we propose to reconstruct the interior of the CFRP samples with the Neural Network Filtered Backprojection (NN-FBP) method. This method combines several FBP reconstructions with different filters and is therefore able to reconstruct images in a fast way based on fewer projections with a similar image quality. Results show that the NN-FBP allows to drastically reduce the number of required projection images while still maintaining high quality image reconstructions.