Research And Application Of Pavement Crack Detection Based On Deep Learning And Multi-Scale Analysis

In recent years,with the rapid development of the economy and the advancement of science and technology,On the one hand,people travel more conveniently.On the other hand,the health of roads is getting worse due to the aggravation of the car,bad weather,natural aging and other factors.Cracks are common road diseases that affect road performance.Traditional manual detection methods are not only time-consuming,labor-intensive,low-accuracy,and low in safety.Therefore,the research of automatic crack detection and identification system is of great significance to ensure the safety of traffic.At present,many scholars have done research on automatic crack detection and identification.However,the accuracy of the current pavement crack identification is generally low due to the unevenness of the light intensity of the road surface crack,the topological complexity of the crack,and the noisy texture background,and because the calculation time complexity is mostly high,it is impossible to detect the crack in time,and it is difficult to meet the actual requirements.In view of the shortcomings of traditional manual detection methods and the problems in traditional image processing crack identification,this paper carried out the following research work to improve the crack recognition rate and detect cracks in real time:(1)The traditional crack detection usually uses on-the-spot photographing and offline identification to detect cracks.In order to realize real-time on-line detection of pavement cracks,this paper studies and implements the pavement crack detection based on deep learning.(2)In order to realize the rapid detection and recognition of pavement crack images,a deep learning network based on YOLO v3 is used to detect pavement cracks.On this basis,in order to realize the detection of crack video,OpenCV is used to frame the video and identify it with the model,then the video frame is restored to video,thus the real-time detection of crack is realized.(3)Since the collected crack images occasionally have noise such as background texture,it is difficult for the model to detect cracks,and the data set of such crack images is very small,and model training cannot be used to improve the detection accuracy of such cracks.In response to this situation,multi-scale methods are used to process image noise.Firstly,the image is processed by non-subsampled Contourlet transform.The noise signal is removed according to the different frequency domain features of the crack and the background texture,and then the image is reconstructed by the inverse non-subsampled Contourlet transform to obtain the denoised image.Finally,the crack image after denoising is detected and identified.(4)The software part of the pavement crack detection system is researched and designed,including user interface module,video segmentation module,crack detection module,database module,etc.And each module of the system is tested.The test results show that the system can complete the detection of pavement cracks and achieve the expected results.