Super-resolution reconstruction of vertebrate microfossil computed tomography images based on deep learning

Micropaleontologists use the fine structures of microfossils to extract evolutionary information. These structures could not be directly observed with the naked eye. Recently, paleontologists resort to computed tomography (CT) images to mine the information, and pursue higher resolution CT images with in-depth research. Therefore, we propose a new model, weighted super-resolution generative adversarial network (WSRGAN), for the super-resolution reconstruction of CT images. The model proposed herein (WSRGAN) obtained higher LPIPS (0.0757) on the experimental dataset, compared with Bilinear (0.4289), Bicubic (0.4166), EDSR (0.2281), WDSR (0.2640), and SRGAN (0.0815). WSRGAN meets the requirements of paleontologists for reconstructing fish microfossils. We hope that more super-resolution reconstruction methods based on deep learning could be applied to paleontology and achieve better performance.