A Projection Error Optimization Neural Network Based Sparse CT Reconstruction Method for Carbon/Carbon Materials

In ⁶⁰Co based computed tomography (CT) of carbon components, reducing the number of sampling angles can significantly shorten detection time and improve detection efficiency. However, for conventional analytical reconstruction algorithms, sparse angle reconstruction images contain a large amount of noise and artifacts, which interfere with the detection of defects in the images and affect the quality evaluation of the inspected components by the detection system under fast detection conditions. This article proposes a sparse angle CT image reconstruction method based on neuralnetwork, which uses an untrained encoding decoding neural network to optimize the projection error of the reconstructed image, and uses the ADAM algorithm to optimize the total variation prior of the image. Compared with traditional deep learning reconstruction algorithms, this method does not require training sample sets and has stronger generalization ability and robustness. The results of simulation and practical experiments show that compared to traditional analytical and reconstruction algorithms, this method significantly improves the quality of reconstructed images, while retaining the detailed information of the detected components, and significantly suppresses artifacts and noise in the reconstructed images retaining image details and texture. This work can effectively improve image quality, eliminate the interference of artifacts on the identification of defects in graphics, and improve the ability to recognize defects in carbon components for the detection system.