AutoWindLoc: Precise Localization of Wind Turbines in High-Resolution Orthophotos for Enhanced Registers

The paper proposes a novel framework for automatically detecting wind turbines in orthophotos, transferring this information to a database, and linking detected turbines to an existing registry to minimize location inaccuracy. This inaccuracy has a significant impact on planning and identifying new potential locations for wind turbines, as existing turbines must be considered in these processes. The existing public data frequently exhibit discrepancies from the actual location, and existing work also exhibits relatively large discrepancies from the actual location, even though the exact location of a wind turbine is so important for these processes. Moreover, existing work has not produced a new or improved database that could be used in the long term for processes in the wind energy sector. The development of the AutoWindLoc framework creates a fully automated data basis from which locations and possible further information can be retrieved. The recognition process utilizes a two-stage approach, incorporating a You Only Look Once model with negative sampling and a binary classification Convolutional Neural Network, which attains an average deviation of 0.85m from the actual location.