November 25, 2026
Europe/Bratislava timezone

Automatic Classification of AMOS Composite Images Using YOLO Models

Not scheduled
1m
Vestibul pavilónu Fyziky

Vestibul pavilónu Fyziky

Zamestnanci Fyzika Poster session + coffee: works of research groups of the faculty Physics Poster session + káva: prezentácie vedeckých výsledkov FMFI UK - Fyzika

Speaker

Karol Havrila (Vedecký pracovník, Astronomické a geofyzikálne observatórium Modra)

Description

The All-sky Meteor Orbit System (AMOS) records meteors and a wide variety of non-meteor events. The growing volume of composite images makes manual inspection and classification time-consuming. We use YOLO models to automate image classification and object detection. The current YOLO11n model detects meteors and classifies each image according to the presence of a meteor detection. On an independent test set of 747 images, it achieves 97.46% classification accuracy and 97.29% recall, missing 13 of 480 meteor images. Further development combines image-level classification, object detection and derived labels in a common processing pipeline. The aim is to reduce manual work and enable automated processing directly at AMOS observing stations.

Pracovisko fakulty (katedra)/ Department of Faculty Astronomické a geofyzikálne observatórium Modra, OAA, Katedra Astronómie, fyziky Zeme a meteorológie
Tlač postru/ Print poster Budem požadovať tlač /I hereby required to print the poster in faculty

Author

Karol Havrila (Vedecký pracovník, Astronomické a geofyzikálne observatórium Modra)

Co-authors

Mr Samuel Fodor Juraj Toth (Katedra astronómie, fyziky Zeme a meteorológie FMFI UK) Tomas Paulech (Faculty of Mathematics, Physics and Informatics, Comenius University in Bratislava) Viera Krešnáková Peter Butka

Presentation materials

There are no materials yet.