针对公路路面裂纹的计算机视觉安全检测方法

Computer Vision Security Detection Method for Highway Pavement Crack

以江苏省南京市竹山路公路养护工程为例,对公路裂纹病害问题的计算机视觉安全检测方法展开研究。提出基于半监督训练的融合显著性半裂纹检测方法,分别从整体设计思路和具体实现方法两方面说明其实现路径,并根据RCD数据集设计检测试验结果显示,提出的检测方法精度为83.94%、召回率为95.08%、F1-score为88.52%,说明该方法能显著降低对手工标注信息的依赖,具有较高的检测精度、抗噪声能力和鲁棒性。公路养护工程实际裂纹检测效果分析证明,此方法能够实现绝大部分的真实图像还原,应用效果优于语义分割网络和裂纹检测网络,在细小裂纹、边缘处仍存在小范围的漏检和误检现象,还存在进一步优化的空间。

Taking the highway maintenance project of Zhushan Road in Nanjing city,Jiangsu province as an example,the computer vision safety detection method of highway crack disease problem is studied.Fusion significant half crack detection method based on semi-supervised training,respectively from the overall design idea and specific implementation method,and according to the RCD data set design test results,the proposed detection method accuracy is 83.94%,the recall rate is 95.08%,F1-score is 88.52%,indicating that the method can significantly reduce the dependence on manual annotation information,has a high detection accuracy,anti-noise ability and robustness.The analysis of the actual crack detection effect of highway maintenance engineering proves that this method can realize most of the real image reduction,and the application effect is better than the semantic segmentation network and crack detection network.There are still small areas of omission and false detection at small cracks and edges,and there is still room for further optimization.