Investigating Artificial Neural Networks for Detecting Aircraft Wake Vortices in Lidar Measurements

Kurzfassung

Convolutional Neural Networks (CNNs) are employed to identify wake vortices via their two-dimensional position and circulation strength in Light Detection and Ranging (lidar) measurement scans. A campaign at Vienna International Airport delivered data that so far has only been processed with a traditional lidar processing algorithm, namely the Radial Velocity (RV) method. Its not fully automated nature led to only a fraction of scans from the overall data set to be evaluated. Here we present ways to use CNNs for this task. A scoring algorithm engineered for verifying CNN detections has been implemented. In particular green detections (those marked as correct CNN detections by the scoring algorithm) can confidently be used for further analysis about the wake vortex encounter hazard. With this approach we end up with a significantly more processed and characterized lidar data compared to that so far delivered by the RV method.