Deep-learning-based icing identification method for unmanned aerial vehicle

Icing on the wings of an aircraft increases drag and reduces lift, leading to a decrease in the aerodynamic performance and stability margin of UAV, which affects flight security. In order to solve the problem of icing state identification of UAV, this paper establishes a simulation model of icing on the wing of a certain type of UAV, analyses the impact of wing icing on the dynamic response of UAV, conducts simulation experiments with multiple state points and different degrees of icing, establishes a deep-learning neural network for icing state identification, and carries out the training and testing of three different strategies based on the simulation data. The results show that the icing identification method proposed in this paper has high accuracy in the prediction of icing degree and can provide a reference for the design of natural icing degree identification and prediction system for UAV.