Two ResNet Mini Architectures for Aircraft Wake Vortex Identification

Abstract

The identification of aircraft wake vortex is an essential issue in the operation of airspace utilization ratio. In particular, accurately identifying wake vortex in fine classification is helpful to guide separation standards under realistic airport conditions that consist of various complex operation scenarios. To stress this issue and improve the efficiency at the same time, we developed two mini architectures with each network of 10 layers by modifying deep residual neural network (ResNet) and describe the results of a study to evaluate the performances for identifying wake vortex in fine classification. For this purpose, we built the wake vortex dataset measured with pulsed Doppler LiDAR at Chengdu Shuangliu International Airport from Aug 16, 2018, to Oct 10, 2018. To support wake vortex identification in fine classification, the classification indices that consider the background wind speeds, wake vortex evaluation and aircraft types were included in the learning and identification tasks. We compared the performance of the two ResNet mini architectures with other lightweight networks by using wake vortex dataset. The experimental results demonstrate that the developed two ResNet mini architectures contribute to competitive wake identification modeling in terms of accuracy and parameter number.