The data were processed for the training of AI models, so that after quality control a total of approx. 680,000 images were available for training the AI models for object detection. This included about 420,000 images of birds, which in turn were manually annotated by ornithologists according to species and species groups. This resulted in about 72,000 images for the kite species group. With the images annotated by ornithologists, further AI models were trained to recognise the species group "kite".
The three-year research project BirdRecorder, which was funded by the BfN with funds from the BMUV, aimed to develop and test a system for avoiding potential impacts on birds from wind energy use. The BirdRecorder thus represents an anti-collision system for bird-friendly operation of wind turbines. The target species group for the detection of birds with artificial intelligence (AI) methods for the BirdRecorder system are red and black kites.
BfN publication 651 BirdRecorder (2023)