Abstract: Deep learning technology has been widely applied across various engineering fields. To achieve accurate and efficient detection of concrete cracks, an enhanced reinforced concrete crack detection network model—named YOLOv8-Crack Damage (YOLOv8-CD)—was proposed based on an analysis of the strengths and limitations of the conventional YOLOv8l algorithm. First, the RepGhost module was introduced into the original YOLOv8l model, reducing both computational load and the number of parameters. Second, a convolutional attention mechanism (CBAM) was incorporated, enabling the model to focus more effectively on crack regions and thereby improving the accuracy and robustness of concrete crack detection. Finally, the Global Feature Pyramid Network (GFPN) and an optimized loss function were integrated to enhance the comprehensiveness of target feature representation. Engineering application results demonstrated that the improved YOLOv8-CD model achieved higher accuracy, with a mean average precision of 96.4%, an increase of approximately 23.4 percentage points over the baseline model. Additionally, the enhanced model effectively gains a balance between detection accuracy and real-time performance.