Abstract: Seismic exploration is one of the most important technologies in global oil and gas exploration. However, due to complex environments and economic constraints, acquired seismic data often contain missing traces, which pose challenges for subsequent seismic signal processing. In recent years, benefiting from the excellent feature extraction capabilities of Convolutional Neural Networks (CNNs), CNN-based methods have been successfully applied to seismic data reconstruction. Nevertheless, CNN-based methods mainly focus on local features and often neglect global information. In addition, supervised learning methods require high-quality paired training data, which poses a challenge for 3D seismic data. To address these issues, we propose a self-supervised reconstruction network, STC-Net, which integrates Swin Transformer and CNN (STC) to extract and fuse global and local features for 3D seismic data reconstruction. This study establishes a complete workflow, covering the analysis of CNN feature extraction capability, the fusion strategy of local and global features, and the performance evaluation of 3D seismic data reconstruction. Results demonstrate that, compared with conventional methods, STC-Net achieves superior performance in reconstructing 3D seismic data.