Research On Pavement Crack Detection And Segmentation Methods Based On Deep Learning

Highway pavement cracks are common pavement distresses in highway maintenance management.Accurate and efficient detection of road surface cracks is the key to highway maintenance.At present,the method based on manual detection of highway cracks is inefficient,inaccurate and has potential safety hazards.With the rapid development of science and technology,deep learning technology has been widely used in pavement image detection.Based on the analysis of the types and characteristics of asphalt pavement crack images,this paper proposes an improved algorithm based on YOLOv5 s for pavement crack image detection and classification,and road crack image segmentation based on Res2Unet-CBAM network model.The main work is as follows:(1)The image preprocessing work is carried out for various types of asphalt pavement crack images.The preprocessing can eliminate the irrelevant information in the image and restore the useful real information.including image grayscale processing,image denoising and image enhancement;Traditional image segmentation algorithms,such as edge detection,threshold segmentation and morphological processing,are analyzed.Finally,feature extraction is performed on the resulting image samples to obtain the feature information of the image.(2)The common target detection algorithm model is analyzed.The task of target detection is to find all interested targets in the image and determine their categories and positions.The target detection algorithm model includes R-CNN series,SSD and YOLO series.Finally,according to the characteristics of the crack image,YOLOv5 s algorithm is selected and improved.The experimental results show that the improved YOLOv5 s pavement image classification algorithm can accurately select the crack location under complex background,and realize the location recognition and classification of the crack target.(3)A variety of semantic segmentation models based on deep learning are analyzed.The purpose of semantic segmentation is to assign each pixel of an image to a meaningful category in order to achieve higher precision image analysis.Including Unet,Res2 net and other models,combined with Res2 net multi-scale feature extraction module and UNET network structure,Res2-UNET multi-scale pavement crack segmentation network model is designed.Based on the analysis of multiple channel attention mechanisms(including SE,ECA,CBAM modules),the CBAM channel attention module is introduced into Res2 Unet,and a new Res2Unet-CBAM network model is constructed for image segmentation tasks.The Res2Unet-CBAM model was compared with other deep learning models after 100 rounds of training with the augmented dataset.The results show that this model has better image segmentation effect.Finally,the damage degree of pavement cracks is quantitatively evaluated with reference to the specifications,and corresponding maintenance suggestions are given.