Unmanned aerial vehicle application in civil structures monitoring via deep learning

This paper provides a deep learning-based methods for civil structures monitoring in aerial imagery from unmanned aerial vehicles (UAVs). This algorithm is prepared to increase the level of structures monitoring when combined with UAVs techniques. The case study presented herein is the use of drones to monitor the cracks in concrete structures. The structures monitoring using existing UAV method is mainly aimed at the extraction of structures state maps. At the same time, the influence of UAV self-motion on the detection accuracy of cracks has not been overcome, and it is difficult to meet the application of intelligent monitoring system for concert crack detection. Therefore a concrete crack detection using UAV perception method based on UAV imaging and deep learning algorithm fusion is proposed. First, the development architecture of deep convolutional neural network (CNN) in the field of computer vision for UAV generated image processing are introduced. Then, the existing crack image datasets and model performance evaluation metrics are summarized. Finally, UAV platform for crack detection in civil infrastructure are summarized. The results show that the proposed CNN network has achieved better performance, reaching 96.4%, 93.7%, 91.3%, and 125s in terms of accuracy, regression, F1-score, and training time, respectively, which can realize the automatic extraction of high-dimensional, complex, and abstract features of civil structures, in addition, the detection results can reduce interference, reduce detection error, obtain more completion, and clear the civil structures detection effect.