Research On Pavement Crack Detection Based On 2D And 3D Fusion Image Recognition And Segmentation Double Layer Deep Learning Algorithm

Cracks,as damage diseases with large volume on the pavement,are easy to develop into other diseases if not timely detected and repaired,thus affecting the service life of the road.At the same time,pavement cracks are slender in form and small in area,accounting for only a small part of the overall road area.In order to detect the crack area in a targeted way and improve the detection efficiency and accuracy,This paper proposes a road crack detection method based on the recognition and segmentation two-layer deep learning algorithm.The specific research content is as follows:(1)Based on the pavement image data collected by the road multifunctional detection vehicle,this paper preprocesses the original image to enhance the image information through brightness transformation,image fusion and image downsize,and makes the image classification,target recognition and image segmentation data sets of pavement cracks respectively.(2)In this paper,Mobilenet V2 and YOLOv5 models were used for image classification and target recognition of pavement crack images respectively.The input images were processed by pixel-level fusion to improve the detection ability of the model,and the influence of pavement images of different scales on the pre-screening results of the crack area was studied.The transfer learning method used in classification model is improved,and Bi FPN module and Eiou loss are introduced into target recognition model to improve their ability to identify crack areas.(3)Based on the U-net model,this paper introduces residual structure in the encoder part to deepen the model level,and introduces deep deseparable convolution,batch normalization layer and 1×1 convolution check model in the decoder part to improve the lightweight and enhance the information interaction between channels.Based on the improved U-net model,the influence of different source and different scale input images on the segmentation results of the model is studied.The results show that the detection accuracy and efficiency of the proposed method are improved compared with those before the improvement.F1 index is increased by3.41%,m Io U index is increased by 1.88%,and the detection speed reaches 59.54 FPS.(4)This paper proposes a pavement crack detection method based on the recognition-segmentation double-layer deep learning algorithm,designs the recognition-segmentation pavement crack detection method based on image classification and target recognition,calculates the crack geometric parameter information based on the segmentation binary graph,and integrates the above detection modules into a set of detection system platform.The visualization and information export of crack detection process are realized.Experimental results show that the pavement crack detection method proposed in this paper based on the recognition and segmentation two-layer deep learning algorithm reaches 82.72%and 84.27% in F1 and m Io U values,and 110.53 in FPS values,respectively,which has a better detection effect,and can provide a certain reference for the subsequent decision-making of pavement maintenance.