Research On Concrete Crack Detection And Quantification Method Based On Depth Learning

The mileage of highway construction in China has increased year by year,up to 5.28 million kilometers,and the maintenance ratio has reached 99.1%.As the mainstream disease in current road maintenance,the rapid and accurate detection and quantification of cracks is helpful to make more accurate decisions for intelligent road maintenance.However,the manual detection efficiency is low,and the traditional crack disease detection algorithm is generally poor,which is difficult to adapt to the actual road environment changes.Therefore,using more intelligent methods to detect and evaluate road conditions has become a current research hotspot.Based on deep learning,this study constructs a U-shaped multi-scale crack detection network to adapt to the multi-road complex environment of crack detection and realize the automatic detection of cracks,which provides a feasible method for road automatic maintenance decision-making.The main research contents are as follows:(1)Construct a pixel-level crack detection training data setWhen using supervised learning methods,rich sample data can effectively improve the robustness and accuracy of the model.Therefore,based on the collection of public data sets and self-collected data,a variety of data amplification methods are used to expand the training data set.At the same time,aiming at the problem of time-consuming and labor-intensive manual labeling,a semi-automatic crack labeling process is proposed to improve the labeling efficiency.Moreover,after analyzing the noise interference factors of the crack image,a variety of noise reduction methods are proposed,including filtering methods and a shadow removal method.(2)Multi-scale dilated convolutional attention encoding-decoding network based on deep learningAfter understanding the basic principles of deep learning and analyzing the classical image segmentation network structure,a multi-scale dilated convolutional attention encoding-decoding network is proposed to extract crack pixels in road images.Firstly,aiming at the problem of complex crack topology structure,a multi-scale void convolution structure is proposed to make the network more fully extract crack features.At the same time,pyramid attention and channel attention are introduced to weaken background noise interference and improve the robustness of the model to crack detection in complex environments.Finally,through the mixed attention mechanism,the low-level features and high-level features of the cracks are fused,so that the crack binarization results are more refined.(3)Fracture detection feature quantization and interaction systemAfter obtaining the binary image of the crack,the geometric characteristics of the crack are quantitatively calculated,and reasonable repair suggestions are given.Using GUI technology,the operation is encapsulated,and a set of crack detection interactive system is built,which provides an efficient and intelligent method for road maintenance decision.