Research Of Asphalt Pavement Crack Detection Methods Based On Deep Learning

China’s intelligent transportation system is in a critical period of construction.The construction of intelligent transportation system is of great significance to the management and maintenance of basic road facilities and road traffic environment.Realizing high-quality maintenance management of road facilities is one of the tasks of the intelligent transportation system.The timely detection and repair of road surface diseases is an important task of road maintenance.Asphalt material due to its durability is widely used in road construction,based on the collected asphalt pavement images,cracks are the most common and most important manifestation of these diseases,and realize high accurate pavement cracks recognition is the guarantee of achieving high quality road maintenance management,also is the important foundation of construction of intelligent transportation system.At present,the method based on digital image processing is the mainstream road cracks detection algorithm,but due to the influence of the environment and equipment at the time of acquisition,the quality of the images collected in the actual environment are uneven,resulting in that these algorithms can’t meet the actual engineering needs.In recent years,deep learning has been widely used in the field of image recognition and achieved good results.Based on the research status of asphalt pavement cracks detection technology,this dissertation analyzes the characteristics of asphalt pavement images collected in the actual environment,and conducts a systematic study of asphalt pavement cracks detection methods based on deep learning.The main contents of the dissertation are as follows:(1)In order to reduce the influence of noise,obstacles and uneven brightness on the detection results of the asphalt pavement images collected in the actual environment,the methods for preprocessing the asphalt pavement images are first studied.The gray level threshold segmentation is used to remove the road marking line in the road images.The adaptive median filter method is used to to filter the noise in the road images as much as possible,and the histogram equalization method enhances the contrast between the cracks and the background.(2)Aiming at the irregular shaped cracks such as lumps and hiatus in the images of asphalt pavement,the dissertation proposes pavement cracks detection models CRes-Net and ACRes-Net.CRes-Net is a cracks detection model after optimizing the residual neural network using the network layer recombination method,which has a certain improvement in the detection performance of cracks.ACRes-Net is a new crack detection model based on CResNet,after introducing the attention mechanism,it can learn the characteristics of crack targets more fully,thereby further improving the detection accuracy of the model.(3)Aiming at the regular shaped cracks such as horizontal or vertical in the images of asphalt pavement,a pavement cracks detection model is proposed based on improved Faster R-CNN.Based on the classic object detection model,Faster R-CNN,its feature extraction network and region network are improved according to the particularity of the crack detection task,thereby improving the model’s detection capabilities.