GCI: A generative CT interpolation model for rapid wood CT scanning

Micro-computed tomography (micro-CT) is a key technique for acquiring three-dimensional images of materials, yet its long scanning and reconstruction time and high cost limit large-scale application. Fast scanning by varying pitch value to loss of image information. This paper proposes a generative CT interpolation model named GCI, which recovers missing information resulting from a fast scanning strategy, enabling fast and high‑quality 3D imaging of wood. GCI is built on a 3D ResNet‑18 backbone, incorporates a 3D CBAM attention mechanism, and adopts an encoder‑decoder architecture to achieve multi‑scale feature fusion, thereby avoiding artifacts and blurring caused by traditional optical‑flow methods under complex structural variations. This study also constructs the first dedicated dataset for wood CT Z‑axis super‑resolution, named CT‑WOOD. Experiments show that GCI outperforms existing mainstream interpolation methods in metrics such as PSNR, SSIM, and LPIPS, achieving a 22% improvement in perceptual similarity (LPIPS) over the second‑best model. Visual comparisons and segmentation experiments further verify the superiority of GCI in detail restoration and structural preservation. When using this fast scanning strategy to save time by up to a factor of four, GCI can reconstruct the image information lost due to the strategy, thereby restoring image quality. This provides a reliable technical pathway for efficient three‑dimensional characterization of wood and other biomass materials.