基于YOLOv4的隧道表面病害检测算法

A Tunnel Surface Diseases Detection Algorithm Based on YOLOv4

隧道病害的及时发现与维护对行车安全非常重要,但隧道环境复杂多变、表面图像对比度低,传统模式识别方法无法有效检测病害。对此,文章提出一种基于YOLOv4的隧道表面病害检测方法,其首先采用CSPDarknet-53作为主干网络来有效提取特征,并通过空间金字塔池化(SPP)融合不同尺度特征,然后经过YOLO层分类与回归病害区域,最后应用CIoU损失函数计算回归损失,有效提高了检测精度。试验结果表明,采用该算法及NVIDIA GeForce 2080Ti显卡,检测速度可达到55帧/s;在所建立的高速铁路隧道表面图像数据集中,平均精度均值(mAP)达到65.1%,缺陷检出率达到90.1%,验证了该算法的高效性。

Timely detection and maintenance of tunnel diseases are very important for traffic safety.Aiming at the problem that tunnel environment is complex and variable and the contrast of surface image is low,which makes the traditional pattern recognition method cannot effectively detect diseases.In this paper,a tunnel surface disease detection method based on YOLOv4 is proposed.CSPDarknet-53 is used as the backbone network to extract features effectively,different scale features are integrated through SPP to classify and regress the disease area through YOLO layer,and the detection accuracy is effectively improved by using CIoU to calculate the regression loss.Experimental results show that the detection speed can reach 55fps by using this algorithm and NVIDIA GeForce 2080Ti.In the established surface image data set of high speed railway tunnel,the mAP reaches 65.1%,and the defect detection rate is 90.1%,which verifies the high efficiency of the algorithm.