基于改进YOLOv9的路面病害检测模型

打开文本图片集

中图分类号：TB9；TP391.41；U418.6 文献标志码：A 文章编号：1674-5124（2025）07-0019-11

Abstract： Aiming at the problems of large size differences of existing pavement diseases and the diffculty in extracting the features of fine diseases in the pavement disease detection task，based on the YOLOv9 network model，this paper first introduces the intra-scale feature interaction AIFI module to provide more comprehensive informationunderstandingand deeper feature extraction.Secondly，thecross-scale feature CCFF module is introduced to improve the adaptability ofthe model to the variation of the target size; Finally， the Focaler-IoU boundary regresson loss function is introduced to reduce the influence of the distribution of difficult samples and manageable samples on the bounding box regression. This method conducted multiple sets of experiments on the Chinese regional dataset of RDD2022. The experimental results show that compared with the original YOLOv9 model， the improved model has a 3.3% increase in average accuracy，a 3.5% （204号 increase in accuracy，and a 4.6% increase in recall rate with little frame rate loss.A series of experimental results show that the method proposed in this paper has a beter detection effect in the task of pavement disease detection.

Keywords：pavement distress detection;cross-scale feature;YOLOv9

0 引言

公路是覆盖范围最广且最常见的交通基础设施在日常生活中为人们的出行提供了便捷的服务。（剩余15800字）