Abstract: Road object detection experiments in complex road environments are prone to missed detections, false alarms, and repeated detections. To address real-time road damage detection under complex road conditions, this paper proposes an improved algorithm based on YOLOv11. An efficient novel grouped extraction convolution is designed to replace the convolutions in the feature extraction network of the original YOLOv11n model, significantly expanding the model's receptive field and enhancing its detection capability for small-scale targets. To tackle issues such as weak illumination relationships, low light, and shadow overlaps in complex road scenarios, the upsampling method is improved by using the novel DySample upsampling operator to enhance the contrast between targets and the background, while effectively improving the localization accuracy for densely distributed road distress. Finally, the C2DyBRA module is introduced to strengthen the model's ability to capture feature details, thereby optimizing the accuracy and matching degree of bounding box predictions. Experimental results show that the improved algorithm, while maintaining the lightweight nature of the original baseline model, achieves a 9.8% increase in mAP@0.5 and a 7.7% improvement in recall compared to the original baseline model. It significantly enhances detection accuracy while meeting real-time requirements, demonstrating superior detection performance.