Research On The Segmentation Method Of Airport Pavement Cracks Based On Deep Learning

With the continuous acceleration of airport construction,the pavement surface monitoring environment has become increasingly complex,maintenance work has become more onerous,and safety issues have become increasingly prominent.In order to ensure the safety of aircraft and passengers,it is necessary to pay attention to the structural safety of the road surface.However,the traditional pavement crack detection method is still based on manual visual inspection,which is low in efficiency,poor in accuracy and high in cost.In order to meet the requirements of the current rapid development of airport construction,it is urgent to use efficient and accurate road surface inspection tools.Proposing an automated road surface detection method is of great significance to airport road surface safety.Therefore,this article is divided into the following three aspects for the research of airport pavement cracks:1.Aiming at the currently undisclosed airport pavement crack data set,this paper uses pavement detection robots to collect real pavement data from multiple domestic airports,manually screening diseases,analyzing the causes of cracks,manually marking the diseases,and constructing the data set.Due to the small amount of data,data amplification is carried out through image translation,rotation,flipping and crack image pasting to form a large-scale airport pavement crack data set.2.In view of the problem that the existing algorithms can not extract the airport pavement cracks accurately,a new algorithm based on deformable convolution and feature fusion is proposed.This model takes into account the characteristics of cracks with diverse shapes,different sizes,and unique topological structures.First,deformable convolution module is used to extract the characteristics of the cracks,and then a multi-scale module composed of different sizes of convolution kernels obtains different receptive field under the scale,and finally by fusing the low-level and high-level features of the crack to obtain a more accurate segmentation result of the crack.The experimental results show that the algorithm can obtain a high accuracy rate for the detection of cracks on the airport pavement.3.When small cracks are hidden in complex backgrounds such as water stains,asphalt,repair blocks,and anti-skid lines,the above algorithms have the phenomenon of under-segmentation.Therefore,in order to solve the problem that cracks are difficult to completely segment under complex backgrounds,a detection algorithm for small cracks in airport pavements under complex backgrounds is proposed.The encoder part of the network uses VGG19 as the backbone network to extract crack feature information,and pass the extracted crack information to the spatial pyramid pooling module to obtain global crack information..By introducing hole convolution and multiplying at different stages of the decoder Loss supervision function to obtain a larger receptive field to improve the effect of crack segmentation under complex background.Experimental results show that the network model is superior to crack segmentation in complex backgrounds,and at the same time it has a faster speed than the above algorithms.