Rock CT Image Segmentation Based on Transfer Learning and U-Net

In this study, we delve into the performance nuances of U-Net and its variants through transfer learning in rock image segmentation. Employing a robust rock image dataset alongside standard evaluation metrics, our analysis unveils the distinctive capabilities of each network variant. The experimental outcomes reveal the diverse efficacy of these networks in segmentation tasks, guiding the selection process based on specific application needs. Particularly, Swin-Unet emerges as a balanced option for achieving high accuracy with computational efficiency, while AttUnet demonstrates superiority in processing complex textured rock images. These nuanced recommendations aim to aid researchers in choosing the most fitting network for enhancing segmentation precision and efficiency in rock image analysis, offering valuable insights for both academic and practical advancements in this domain.