In order to enhance the intelligence level of shale gas reservoir fracture identification and achieve automated analysis of electrical imaging logging data, an integrated deep learning approach was used to investigate fracture characterization in the Wufeng-Longmaxi Formation shale reservoir. The methodology combined MobileNetV3's lightweight architecture with Swin Transformer's attention mechanism, incorporating Watershed position encoding and AFPN-CA multi-scale fusion. The results show that the model excels in fracture identification for the Wufeng-Longmaxi shale reservoir, achieving 92.3% intersection accuracy—a 25% improvement over traditional methods. Filtersim-Inpainting preprocessing boosts image quality with 98.6% repair completeness, while maintaining 45 FPS processing speed and 20× efficiency gains in well interpretation. It is concluded that this innovative integration of geological priors with deep learning features enables reliable, automated evaluation of complex fracture networks, supporting intelligent analysis of four distinct fracture types and their geometric parameters in shale reservoirs.