一种新的沥青路面灌封裂缝自动提取方法  

A Novel Approach for Automatic Extraction Asphalt Pavement Sealed Crack

作　　者：邓砚学 张志华 张新秀[4] DENG Yan-xue;ZHANG Zhi-hua;ZHANG Xin-xiu(Faculty of Geomatics,Lanzhou Jiaotong University, Lanzhou 730070, China;National-Local Joint Engineering Research Center of Technologies and Applications for National Geographic State Monitoring, Lanzhou 730070, China;Gansu Provincial Engineering Laboratory for National Geographic State Monitoring, Lanzhou 730070, China;Key Laboratory of Highway Network Monitoring in Gansu Province, Lanzhou 730070, China)

机构地区：[1]兰州交通大学测绘与地理信息学院,兰州730070 [2]兰州交通大学地理国情监测技术应用国家地方联合工程研究中心,兰州730070 [3]甘肃省地理国情监测工程实验室,兰州730070 [4]甘肃省公路路网检测重点实验室,兰州730070

出　　处：《科学技术与工程》2022年第16期6687-6694,共8页Science Technology and Engineering

基　　金：国家重点研发计划(2017YFB0504201,2017YFB0504203);国家自然科学基金(41861059,41761082,61862039);兰州交通大学优秀平台(201806)。

摘　　要：为了提高基于图像处理的沥青路面病害识别效率和精度,引入图像增强处理中的多尺度视网膜(multi-scale Retinex,MSR)算法以减弱光照不均匀、道路场景多变等因素对路面病害图像质量的影响。针对SegNet网络难以精确分割沥青路面微小病害的问题,采用比视觉几何群网络(visual geometry group network,VGG)效果更好的残差网络(residual network,ResNet)作为主干网络,同时加入空洞卷积(dilation convolution)层,提高网络对细小病害的识别性能;针对改进网络在识别病害时误检率较高的问题,运用阈值法剔除分割结果中的假阳性。为了验证改进算法的有效性,将其与具有代表性的语义分割方法(如SegNet、BiSeNet)在相同数据集上进行对比,三者的平均交并比(mean intersection over union,MIoU)分别为0.7763、0.6743、0.6971,三者的F_(1)分数(F_(1)-score,F_(1))分别为0.8999、0.8743、0.8990。运用所提方法对甘肃省部分路段的路面灌封裂缝进行识别,结果与人工检测相比,漏检率为0.09%,误检率为2.49%。实验结果表明:所提方法能够更精确地提取沥青路面灌封裂缝。In order to improve the efficiency and accuracy of asphalt pavement disease recognition based on image processing,the multi-scale retinex(MSR)algorithm belonged to image enhancement processing was utilized to reduce the factors that seriously affect pavement disease image quality,such as uneven illumination and changeable road scenes.To solve the problem that the SegNet was difficult to accurately segment the fine defects on asphalt pavement,the residual network(ResNet)with better effect than visual geometry group network(VGG)was used as the backbone network,simultaneously,the dilated convolutional layers were employed to improve the recognition performance of the improved network for small diseases.Aiming at the issues that the improved network has a high false detection rate when recognizing diseases,the threshold method was used to eliminate false positives in the segmentation results.In order to verify the effectiveness of the improved network,it was compared with the representative semantic segmentation methods(such as SegNet,BiSeNet)on the same dataset,and the mean intersection over union(MIoU)of the three are 0.7763,0.6743,0.6971,and the F_(1) scores(F_(1))of the three are 0.8999,0.8743,0.8990,respectively.The proposed algorithm was used to segment asphalt pavement sealed cracks in some road sections in Gansu Province.Compared with manual detection,the missed detection rate and the false detection rate are 0.09%,2.49%,respectively.The experimental results show that the proposed method can segment the asphalt pavement sealed cracks more accurately.

关 键 词：图像语义分割 阈值分割 沥青路面灌封裂缝 编解码网络 空洞卷积 

分 类 号：U416.2[交通运输工程—道路与铁道工程]