Research On Visual Evaluation Of Asphalt Pavement Damage Condition Based On Typical Cracks

Cracks are a common disease of asphalt pavements.Traditional artificial crack detection and maintenance methods can no longer meet the needs of high-speed development of modern highways.Aiming at typical cracks in asphalt pavement,methods of conducting visual evaluation of pavement in an efficient and accurate manner has become an important research direction.Aiming at the image data of the asphalt pavement collected in the actual traffic environment,this thesis proposes a visual assessment method of asphalt pavement damage based on typical cracks.The main work of this thesis is as follows:(1)This thesis proposes a nested U-Net network based on channel attention improvement,which realizes the pixel-level detection of asphalt pavement cracks,and solves the problem of low detection accuracy of complex cracks and small cracks by existing crack detection methods.A nested U-Net network(UNet++)is used as the backbone feature extraction network to detect pavement cracks.Then the top-level features of the backbone feature extraction network are fused,and the channel attention is used to optimize the fused features.The final output is a pixel-level crack prediction image.The network improves the crack detection performance of asphalt pavements by making full use of the capabilities of the nested U-Net structure in image segmentation tasks.(2)The crack skeleton extraction method is improved based on two methods of boundary conditions and graphics processing,which solves the problem of burrs and abnormal boundary when extracting the crack skeleton of asphalt pavement.One approach is to improve the Zhang-Suen method based on boundary conditions.In the conditional judgment of the sub-iteration,eight-neighbors complement is performed on the boundary pixels,which effectively solves the problem of abnormal boundary skeleton.At the same time,another approach is to improve the Zhang-Suen method based on two-step graphics processing,which can fix the skeleton abnormality while solving the skeleton burr.Before extracting the skeleton,this method performs graphic expansion and boundary filling operations on the crack image,which can smooth the crack edges and reduce burrs,and finally crop and output the skeleton image without abnormal boundaries.(3)This thesis proposes an assessment method for the damage degree of asphalt pavement based on typical cracks.It improves the accuracy of the existing automated pavement damage assessment methods and solves the problems such as the inability to judge the degree of crack damage and insufficient assessment accuracy.This method uses automated means to detect pavement cracks,and combines artificial standards to evaluate pavement damage conditions,which improves the accuracy of automated pavement damage assessment methods.This thesis implements a visual assessment method of pavement damage for typical cracks in asphalt pavement.The pixel accuracy for detecting typical cracks in asphalt pavement can reach 96.86%,and the pavement damage assessment is more accurate with a relative error of 1.06%.This method can accurately complete the pixellevel detection of typical cracks in asphalt pavements,and deepen the automation of visual assessment of pavement damage.