Research on Bridge Crack Detection Method of UAV Imaging Based on Convolutional Neural Network
This paper introduces a bridge crack detection system based on convolutional neural network(CNN),which can collect crack images via unmanned aerial vehicles(UAVs),extract cracks,and accurately work out the maximum width of cracks,thus facilitating crack detection and improving the crack width calculation accuracy.The UAVs of this system are trained to collect images of deck soffit and edges.Firstly,the CNN model is built up to filter deck images,and subsequently,a slide crack detection(SCD)model that is based on CNN is established considering the characteristics of these selected images,to extract images through moving of small sliding windows,and the image extraction efficiency was compared with that of an improved median filtering image denoising algorithm that is developed on the basis of crack image statistical features.The cracks are classified and the maximum crack widths are calculated and the calculated results are compared with field measurements.It is shown that the proposed bridge crack detection system of UAV imaging slightly disturbs crack images and displays more efficient and accurate crack extraction.The errors between the maximum crack widths calculated by the system and the measured results are less than 0.05 mm,meeting the required detection levels.
NETL
NSTL
万方数据
