Abstract: [Objective] Accurate identification of micro/nanoscale pores and fractures is essential for understanding multiphase interactions in rocks. However, traditional segmentation methods have significant limitations in precisely segmenting complex pore-fracture structures, and the accuracy of results from various methods is often inadequately evaluated in practical applications. [Methods] In this study, a nanometer-resolution pore-fracture dataset of tight sandstone was constructed using micro-focus X-ray computed tomography (μCT) technology. We compared the performance of traditional segmentation methods, such as grayscale thresholding and watershed algorithms, with four deep learning methods based on convolutional neural network architectures (UNet, SegNet, DeepLabv3-ResNet50, and DeepLabv3-ResNet101) for pore-fracture feature extraction at the nanometer scale. [Results] The results demonstrate that deep learning methods generally outperform traditional segmentation approaches for the micro/nanoscale pore-fracture CT image dataset. In particular, the UNet model achieved the best performance across multiple evaluation metrics: its Intersection over Union (IoU) and F1-score improved by 18.70% and 16.47%, respectively, compared to traditional methods, while accuracy reached 99.03%. The standard deviations of these metrics (0.012, 0.010, and 0.004, respectively) further indicate high stability and robustness. For complex nanoscale pore-fracture structures, UNet effectively preserved detail continuity and boundary integrity, showcasing its superior fine-detail extraction capability. The UNet-based 3D reconstruction yielded a porosity of 2.408% (compared to the original porosity of 2.785%), and the constructed pore network model (PNM) showed enhanced overall connectivity, validating its advantages in multiscale pore-fracture identification and structural preservation [Conclusion] Compared to traditional segmentation methods, deep learning models demonstrate highly consistent performance in segmenting micro-fractures and pores with their pore network topology, significantly improving the accuracy of porosity, pore throat, and permeability characterization. This advancement provides a critical foundation for the precise identification and modeling of micro-fractures and pores.