Abstract: Rapid and accurate identification of tunnel cracks is crucial for ensuring the operational safety. However, existing methods struggle with low accuracy and high missed detection rates under challenging conditions involving significant scale variations, complex morphologies, and blurred textures. To address these issues, an improved YOLOv11 model integrating multi-scale visual encoding and scale-dynamic loss is proposed. Specifically, an Explicit Visual Center Block (EVCBlock) is introduced to enhance feature representation of multi-scale crack features. Simultaneously, a Scale-based Dynamic Loss (SDLoss) is designed to adaptively adjust detection weights across different scales, thereby improving the model’s robustness against tiny cracks and complex backgrounds. Experimental results demonstrated that the proposed method outperforms the baseline YOLOv11n, achieving a Precision of 84.5%, Recall of 78.6%, mAP@0.5 of 88.1%, and mAP@0.5:0.95 of 62.7%, representing improvements of 3.8% ,3.7% , 4.0% and 4.0%, respectively., The proposed model provides more reliable crack detection in complex tunnel scenarios, demonstrating superior performance over mainstream lightweight models such as YOLOv5n, YOLOv6n, and YOLOv8n.