To improve the intelligence level of fracture identification in shale gas reservoirs and realize automated analysis of electrical imaging logging data, the TransFracNet deep learning model was proposed. First, MobileNetV3 and the Swin Transformer were integrated to combine lightweight convolution with global attention mechanisms. Subsequently, Watershed positional encoding and an AFPN-CA multi-scale feature fusion module were incorporated into the network architecture. Finally, a preprocessing algorithm based on Filtersim-Inpainting was applied to enhance image quality. Results indicate that the model performs well on the test set of the Wufeng-Longmaxi Formation shale reservoir in the Luzhou block. The accuracy of fracture intersection identification reaches 92.3%, representing a 25% improvement compared to traditional methods. Image quality is significantly improved by the preprocessing algorithm, with a fracture restoration completeness of 98.6%. The system maintains a stable processing speed of 45 FPS, and single-well interpretation efficiency is increased by a factor of 20. Four types of fractures, including low-angle high-conductivity fractures and en echelon induced fractures, are supported for identification, along with the intelligent extraction of their geometric parameters. It is concluded that the TransFracNet model demonstrates excellent performance in handling complex geological structures and multi-scale fracture features. Geological prior knowledge is innovatively fused with deep learning features. The research results provide reliable technical support for the evaluation of shale gas reservoirs.