Research On Crack Segmentation Algorithm Based On Dilated Residual Network

With the increasing investment in infrastructure construction such as highways and bridges in China,road maintenance work becomes more and more important.Pavement cracks are a common highway disease,and timely acquisition of road pavement crack information plays an important role in road maintenance.Crack detection based on traditional image processing algorithms has gradually replaced manual crack detection methods,due to its advantages such as fast detection speed and convenient detection.However,traditional image processing methods generally define a feature first,and then further predict or classify such features.The recognition rate of such methods is relatively low,which can no longer meet the engineering requirements of people today.In recent years,deep learning has received a lot of attention from many researchers.Through a large amount of data training,deep learning is very robust to target detection in complex environments,and its accuracy is far beyond traditional algorithms.In particular,the successful application of deep learning in the field of image segmentation,starting from the application of deep learning in crack detection,this paper deeply studies the theoretical characteristics of deep learning networks,and based on this further design is more suitable for the network model of road surface crack detection in complex environments.The main work of the paper is as follows:(1)A pavement crack data set under a real and complex environment was produced,which was called Pavement Crack-Data.Under the guidance of professionals,the pavement crack images were collected,and then the data set was amplified using image processing methods and the images were manually marked.Get the training labels,in order to meet the model requirements of this algorithm in this paper.(2)The comparison algorithm of the core algorithm in this paper is described,which are respectively the crack segmentation algorithm based on the improved Dense Net full convolutional network and the crack segmentation algorithm based on the generation of adversarial networks.The two are applied to the pavement crack images established in this paper.Based on the experimental results,the core algorithm of this paper is derived.(3)Taking the image segmentation of pavement cracks based on the dilated residual network as the research content,firstly,the model and network structure of the dilated residual are designed,then the specific model optimization strategy is given,and the model training is performed on the Tensorflow platform,and then the network model is detected.The obtained segmented images are subjected to morphological feature extraction,and an iterative thinning algorithm is used to extract the skeleton of the crack image,and finally the traditional image algorithm is used to measure the crack morphology information.(4)This article uses the Flask framework to build an online pavement crack detection and analysis system based on the B/S architecture.The system uploads data by the inspectors,and performs image algorithm processing on the data in the background.It includes two formats: image format and video format,and displays the results visually.Thereby,it provides important reference for inspectors and helps to better assess the actual road damage.Therefore,this article will proceed from the above four aspects,deeply integrate crack detection and deep learning theory,design and construct an efficient and accurate pavement crack semantic segmentation model,and use Python’s lightweight web framework Flask to build an online Pavement crack detection system based on B/S architecture.