Toward Automated Aerial Refueling: Automated Visual Aircraft Identification with Convolutional Neural Networks

Robert L. Mash


In the military domain of autonomous aerial refueling operations, automated visual recognition of an approaching aircraft critically supports mission goals. These scholarly articles leverage recent developments in the field of natural image pattern recognition with deep Convolutional Neural Networks (CNNs). The first article reviews the operational details of CNNs, then demonstrates a hyper parameter optimization process. The second investigates advanced forms of data augmentation in terms of image recognition performance. Finally, the third article demonstrates a novel ensemble confidence measure as well as a modified ensemble compression technique which retains a useful confidence measure in a single student network.