ANN estimation model for photogrammetry-based UAV flight planning optimisation

Abstract

Artificial intelligence (AI) is undergoing a ground-breaking period. Recently, AI affects almost every part of human life. Using AI in path planning for Unmanned Aerial Vehicle (UAV) attracts attention as a novel need. The inputs that form the base of UAV use in photogrammetry are UAV Type (UT), Ground Sampling Distance (GSD), Overlap Rates (OR), and Atmospheric Conditions (AC). Input parameters directly impact the UAV's Flight Time (FT) and Battery Status (BS). Weighting and optimizing these parameters are the main ideas of this study. The effects of input values (GSD, OR, UT, AC) on the outputs (BS and FT) were optimized using Artificial Neural Networks (ANN) in this study. For the analysis, results have been produced in which different training algorithms are preferred (Gradient Descent - GD - and Levenberg-Marquardt - LM). The GD algorithm has reached 77.65% accuracy in FT estimation and 80.91% estimation accuracy on normalized data on the BS. Then, the correlation between the produced model and the input parameters and the output parameters was determined, and the weights of the inputs were revealed. As a result, it was determined that the AC parameter has the most significant effect on BS and FT. Also, it has been identified that the normalization process has a considerable impact on optimization.