Water accumulation and ice formation in traffic tunnels pose prominent safety hazards (e.g., reduced road friction, increased traffic accidents) and threaten structural integrity (e.g., damage to waterproof layers and lining structures). Therefore, the intelligent identification of these two hazards is crucial for safeguarding traffic safety and optimizing tunnel maintenance strategies. The intelligent identification system integrates computer vision, deep learning, and multi-source sensor data fusion technologies. Current state-of-the-art practices adopt deep learning models for target segmentation and detection, combined with robust image preprocessing and post-processing techniques. This technology exhibits significant practical application value, and its continuous innovation and development are expected to substantially enhance the level of tunnel safety management and structural durability preservation.