TY - JOUR
AU1 - Yang, Runfeng
AB - The extensive use application of visual perception technology in an unmanned aerial vehicle (UAV) has brought great changes to the application of UAVs in various fields. It is challenging to detect landmark images for UAVs as UAV video-monitoring equipment needs to analyze the whole frame image through complex calculations in real time. Convolutional neural networks (CNNs) have shown advances in the object detection and segmentation fields. However, during UAV flights in different environments, the performance of landmark detection to deteriorate seriously has been caused by the uncertainty of landmark orientation, the diversity of landmark types, and the similarities in different object classes. This paper presents landmark detection of an UAV based on fast region-based convolutional neural networks (Fast R-CNN) adapting to complex and rapid flying environment changes. The proposed model design is improved by optimizing the model’s objective function, which offers higher accuracy and robustness. The proposed system offers detection in real time by utilizing multi-core processors to realize multi-threaded operations more effectively to improve the operation speed of the overall algorithm. Theoretical analysis and experimental results demonstrate landmark recognition with an accuracy of at least 96% to match deployed in UAVs and make correct multiple classifications simultaneously based on Fast R-CNN.
TI - UAV landmark detection on fast region-based CNN
JF - Arabian Journal of Geosciences
DO - 10.1007/s12517-021-07457-w
DA - 2021-06-03
UR - https://www.deepdyve.com/lp/springer-journals/uav-landmark-detection-on-fast-region-based-cnn-7pgkKw7swv
VL - 14
IS - 12
DP - DeepDyve
ER -