TransCNN: A Hybrid Framework for Aircraft Wake Vortex Detection and Safety Interval Assessment

With the rapid development of the aviation industry, airspace environments have become increasingly complex. In addition, fixed-distance wake separation standards, which are constrained by overly conservative spacing intervals and limited adaptability to dynamic conditions, can no longer meet modern operational demands. Therefore, how to perform real-time detection of wake-vortex data using the Doppler LiDAR and achieve dynamic wake-separation reductions has become a crucial challenge in improving airport operational efficiency. To this end, this article proposes a two-branch colearning framework named the TransCNN model, which combines convolutional neural networks (CNNs) and Transformers to align multidimensional learning tasks. The proposed framework includes a multiscale hybrid attention convolution module that enhances the local feature extraction effect in CNN branches. In addition, it includes a global feature fusion module that integrates global information obtained from transformer branches into CNN feature maps. The proposed framework is verified using wake-vortex data obtained by Doppler LiDAR in the approach areas of Qingdao Jiaodong International Airport. The experimental results demonstrate that the proposed TransCNN model excels in LiDAR-based wake-vortex identification, achieving an accuracy of 99.09%, which represents a 12.09% improvement over traditional support vector machine methods. Further, the variation patterns of the wake-vortex dissipation time under different wind field conditions are analyzed, and the superior performance of the TransCNN model in enhancing flight safety and operational efficiency is validated through Markov chain Monte Carlo simulations.