Research On Crack Detection Algorithm Of Concrete Pavement Image

In recent years,with the booming development of logistics industry that aggravating the operational load on all types of highway engineering in the infrastructure,the maintain requirements for roads are greatly increased,especially for the large number of roads built in the early years.It is urgent for researchers to develop a health monitoring system for highway engineering which uses intelligent monitoring equipment to capture images of surface of concrete pavement and then uses deep learning algorithms or machine vision algorithms to detect crack information in the images and help inspectors develop highway engineering maintenance plans.The existing concrete pavement crack detection algorithm does not take into account the interference of pseudo-cracks such as water stains,light shadows,and black spots in the real pavement scene,and in the actual detection process,the algorithm is easy to misjudge such pseudo-cracks as real cracks,resulting in a high false detection rate.Therefore,this paper conducted research on the detection of cracks in highway engineering.Firstly,the shortcomings of existing crack segmentation algorithms for concrete pavements were analyzed,and the crack segmentation algorithm was improved based on full convolutional neural networks.Then the crack extraction algorithm was tested and validated on the collected dataset.Finally,the developed crack edge extraction algorithm based on region growing and gradient threshold segmentation were used to obtain complete edge information of the cracks to assist inspectors in making reasonable highway engineering health monitoring judgements and formulating accurate road maintenance plans.At the beginning of this paper,we introduced the main component modules of convolutional neural networks and the core idea of full convolutional neural networks,as well as analyzed the advantages and disadvantages of common image segmentation algorithms.Subsequently,this paper introduced the full convolutional neural network model that used for concrete pavement image crack detection based on the VGG-16 structure.The network structure was optimally adapted by incorporating a gradient layer module improving the speed of convergence efficiently and a self-attention mechanism module improving global information learning capability.Besides,a negative sample dataset with pseudo-cracks such as leaves,water stains and branches was created,carrying out comparative experimental analysis in terms of both visual judgement and objective metrics.The results showed that with the introduction of the gradient layer module and the self-attention mechanism module,not only did the network training converge faster,but it was also able to segment the cracks in the concrete pavement images more completely and accurately.Finally,in order to obtain the complete edge information of cracks,this paper investigated a pavement crack edge extraction algorithm based on region growing and gradient threshold segmentation.To start with,the algorithm obtained the complete crack edge chain by growing the region based on the significant changes in gradient magnitude and gradient orientation of the crack edge region.Then the redundant pseudo-cracks were removed by a gradient thresholding algorithm,which in turn gave a more accurate crack edge.By comparing with similar methods,the experimental results demonstrated the advantages of the algorithm in this paper in terms of the better extraction effect of cracks,the higher accuracy rate,the stronger noise resistance,and the faster running speed.