针对坑槽和裂缝两种路面病害检测实时性差、准确率低、易误检漏检等问题，提出了一种改进YOLOv9的路面混合病害算法，实现路面裂缝的自动化检测和识别。首先，在骨干网络中引入AKConv （alterable kernel convolution）替换RepNCSPELAN4中的卷积模块，提高网络对不同病害的特征提取能力，有效解决路面病害与背景环境特征难以区分的问题；其次，在检测头中引入了SimAM注意力机制（selective image attention mechanism）和DySample上采样模块，提高网络聚焦特性并增强提取关键特征信息的能力；最后，采用inner-IoU函数优化模型的权重参数，提升对混合样本的学习能力。实验结果表明，改进后的模型与YOLOv9-c相比较，性能有了显著提升，平均精度提升40.17%、召回率提高了15.99%、mAP模型精度提高了20.95%，该优化算法能够更加精准高效的对路面混合病害进行检测，提高了路面病害检测的准确率和泛用性。

Aiming at the problems of poor real-time detection, low accuracy, and false detection and omission of pavement disease detection including hole and crack, an improved algorithm based on YOLOv9 was proposed to resolve the problem. Firstly, AKConv(alterable kernel convolution) was introduced into the backbone network to replace the convolution module in RepNCSPELAN4, which improves the feature extraction ability of the network for different diseases and effectively solve the problem that road disease is difficult to distinguish from background environment features. Secondly, selective image attention mechanism (SimAM) and DySample sampling modules were introduced to focus on the key information in the detection head, and the capability to extract information features was enhanced more efficiently. Finally, the inner-IOU function was used to optimize the weight parameters of the model to improve the learning ability of mixed samples. The experimental comparison between YOLOv9-c and our model showed that the accuracy, recall rate and MAP of the improved model are increased by 40.17%, 15.99% and 20.95% respectively. The performance has been significantly improved, and the detection effect is more accurately and efficiently, and the accuracy and generalization ability of pavement disease detection algorithm are improved.