UAVs in rail damage image diagnostics supported by deep-learning networks

Abstract The article uses images from Unmanned Aerial Vehicles (UAVs) for rail diagnostics. main advantage of such a solution compared to traditional surveys performed with measuring vehicles is the elimination decreased train traffic. authors, in study, limited themselves diagnosis hazardous split defects rails. An algorithm has been proposed detect them an efficiency rate about 81% not less than 6.9% head width. It FCN-8 deep-learning network, implemented Tensorflow environment, extract by image segmentation. Using this type network segmentation increases resistance changes recorded brightness. This fundamental importance case variable conditions recording UAVs. detection these using Python language and OpenCV library. To locate defect, it contour separate together rectangle circumscribed around it. use UAVs artificial intelligence important element novelty presented work.