Identification And Location Of Corrosion And Fatigue Crack Of Steel Structure Surface Based On UAV Image

As an important damage form in large steel structure buildings,corrosion and fatigue cracks pose a serious threat to the normal operation of the structure.It is necessary to carry out long-term structural health monitoring to ensure the safe and stable operation of the structure.The flexible and convenient monitoring method of UAV(Unmanned Aerial Vehicle)can detect the corrosion and fatigue crack in the structure as soon as possible,and avoid the potential safety accident risk of the structure.In this paper,UAV images are used to identify and locate the corrosion and fatigue crack areas on the steel structure surface.The UAV image is segmented by SLIC super-pixel segmentation algorithm.The proposed segmentation ratio index and CNN model recognition results are used to cross verify the segmentation effect.The parameters of super-pixel segmentation algorithm are optimized to realize the pre segmentation of UAV image.According to the segmented super-pixel region,the corrosion segmentation data set and crack segmentation data set are established respectively.The FPN and PANet algorithms are trained with the corrosion segmentation data set,and the corrosion segmentation model with the highest recognition accuracy of 97% is obtained;the Deep Labv3+ model with the fusion of transfer learning and integrated learning method is trained with the crack segmentation data set,and the fatigue crack segmentation model with the highest recognition accuracy of 99% is obtained.The accuracy of the two models is confirmed by the validation set.Using the super-pixel segmentation algorithm and the trained neural network model to identify the overall image of the structure and locate the possible corrosion or crack area;combined with the UAV flight records and the super-pixel segmentation label,the possible defect area is detected to accurately locate the corrosion or fatigue crack area.Through the example of Tianjin progressive bridge and the case of steel crack in laboratory environment,the practical ability of this method is proved.