J. Sprockhoff, U. Durak

This paper discusses the benefits of using ensembles of neural networks for safety-critical tasks in an aircraft. The advantages of increased performance, uncertainty assessment of predictions and redundancy are highlighted. This is done using the example of an AI-based system to detect other aircraft. The system uses a stereo vision approach with two cameras to determine the distance to other aircraft. It is shown how by averaging predictions over an ensemble of five object detectors the detection rate and the accuracy of distance predictions can be improved over a single neural network.

Deutscher Luft- und Raumfahrtkongress 2023, Stuttgart

Deutsche Gesellschaft für Luft- und Raumfahrt - Lilienthal-Oberth e.V., Bonn, 2024

Conference Paper

englisch

21,0 x 29,7 cm, 9 Seiten

urn:nbn:de:101:1-2024022811514689872431

10.25967/610072

Artificial Intelligence, Neural Networks, Computer Vision, Ensemble Learning, Object Detection

Download - Bitte beachten Sie die Nutzungsbedingungen dieses Dokuments: Copyright protected

Sprockhoff, J.; Durak, U. (2024): Neural Network Ensembles for Safety-Critical Object Detection Functions in Aerospace. Deutsche Gesellschaft für Luft- und Raumfahrt - Lilienthal-Oberth e.V.. (Text). https://doi.org/10.25967/610072. urn:nbn:de:101:1-2024022811514689872431.

28.02.2024