Design Detailing of Machine Learning-based Estimation Method of Stable Communication Range for UAV Flight

In order to realize beyond visual line of sight (BVLOS) unmanned aerial vehicle (UAV) flights for agricultural support in mountainous areas, it is necessary to estimate the flight range of UAVs where the stable communication with ground control terminals can be maintained for observing the condition of the flight path. Existing research proposes methods for estimating wireless communication qualities using radio wave propagation simulations based on ray tracing technology. However, when incorporating complex structures of the field such as vegetation in the mountainous areas, the computational load of the ray tracing markedly increases, making it difficult to perform frequent simulations in agricultural areas where terrain structures change continuously. Therefore, this study proposes a new wireless communication quality estimation method that integrates lightweight radio wave propagation simulation and machine learning technology. The proposed method first roughly estimates communication quality using the lightweight simulation that considers only coarse terrain structures such as elevation. And then, the estimation results are applied to a machine learning model trained to predict communication quality incorporating accurate 3D terrain structures. Through field experiments, we demonstrate that utilizing machine learning models enables the accurate prediction of radio wave propagation in the field by reflecting detailed terrain features.