Deep Learning-based Crack Detection Of Concrete Pavement

As the investment in infrastructure such as highways and bridges in China has increased year by year,structural health issues have always been an important part of China's national economy and people's livelihood.However,surface cracks in engineering structures are one of the key indicators for evaluating structural damage and durability.At present,crack detection is mainly manual inspection in practice,and the method has high work cost,high labor intensity and low detection efficiency.Therefore,it is an urgent engineering problem to grasp the road pavement information quickly and timely and realize the automatic detection of structural surface defects.In recent years,deep learning has been rapidly developed in the field of machine vision,and the accuracy of deep learning recognition has surpassed the traditional image recognition algorithm.Compared with traditional algorithms,deep learning does not require manual design features,and can automatically abstract expression based on the original image features.Therefore,in this paper,a method for detecting cracks in concrete structure images by convolutional neural network is designed for concrete crack detection,and it is applied to concrete pavement crack detection.The main contents of this paper include:First,study the basic theory of deep learning.It mainly introduces the basic theory and principle of deep learning convolutional neural network,namely the overall framework of convolutional neural network,structural function and training process,and common target detection methods.Secondly,the depth image learning is used to complete the crack image classification.A concrete road image is captured by a camera,and image preprocessing is performed to create an image classification data set.Based on the AlexNet framework,the structure and hyperparameters are optimized,and the concrete pavement crack detection model is designed and trained.The model can automatically learn the effective features of concrete surface crack image classification and realize automatic classification of concrete crack images with an accuracy of 98.5%.Finally,the target detection of the crack image is achieved.Based on the image classification dataset,the target detection dataset is generated by image annotation.The Faster learning target is used to detect the Faster R-CNN method.The ZF network is used to identify and locate the concrete image with cracks,and an efficient and reliable image is obtained.Real-time detection method for cracks in concrete pavement from coarse to fine.Through the research of this thesis,the automatic detection of structural surface defects is realized,and the informationization development of engineering inspection technology is promoted.