Research on pavement crack detection technology based on improved YOLO11

Accurate identification and analysis of pavement cracks are crucial for extending road lifespan and improving AGV safety. Current research focuses mainly on overall crack identification, with limited attention to crack width detection. This paper explores pavement crack detection using an improved YOLO11 model. Firstly, a pavement crack data set acquisition system was built to collect a total of 1000 pavement crack images. Secondly, the pavement crack detection model based on YOLO11 is constructed and improved. The Dynamic Upsampling module is used to replace the original Upsample of YOLO11, which improves the detection accuracy and makes the resource utilization more efficient. At the same time, the Detection head is replaced by the Segmentation head to analyze the detection results more intuitively. Finally, the experimental results show that compared with YOLO11, the improved YOLO-Dy achieves a 10.3 % increase in Mask-mAP50 value, a 15.7 % increase in Mask-P, a 6.7 % increase in Mask-R, and a 2.7 % increase in Box-P on the self-collected data set. The performance on the public data set crack is also improved. It provides accurate and reliable support for the detection and repair of cracks of different sizes on the pavement and the prediction of the overall service life of the road.