Logo image
Back
U‐Net Based Machine Learning for Identification of Auroras in Space‐Based Images
Journal article   Open access   Peer reviewed

U‐Net Based Machine Learning for Identification of Auroras in Space‐Based Images

Amit Kain, Matthew G. Finley, Andrew D. Howarth, Andrew W. Yau and David M. Miles
Journal of geophysical research : Space physics (2013 - Present), Vol.131(9), e2026JA035329
09/26/2026
DOI: 10.1029/2026JA035329
url
https://doi.org/10.1029/2026JA035329View
Published (Version of record) Open Access

Abstract

We use a Semantic Segmentation based U‐Net model to predict the appearance of various classifications of aurora present in near‐infrared images from the Fast Auroral Imager (FAI) onboard the CASSIOPE satellite. We manually labeled pixels from a set of 4091 FAI near‐infrared images into 7 different classes: no aurora, contaminated, diffuse, discrete, arc, ambiguous, and vignette. The U‐Net architecture was modified by replacing 3 layers and changing the input layer to be compatible with 280 × 256 pixel images. The challenges for auroral classification with machine learning when subclasses overlap with each other are explored. Tags for images were generated by analyzing predictions from the trained model over a series of successive images. Overall, the trained model was able to achieve 52.8% Mean Intersection over Union for unseen data. The use of this model and the resulting tags as a search filter enable fast and simple discovery of aurora in space‐based images. Plain Language Summary The Fast Auroral Imager (FAI) on the CASSIOPE satellite was designed to take pictures of the aurora from space, and it has taken over 1.7 million images since the satellite was launched in 2013. We have developed an algorithm based on a U‐Net machine learning model, trained on 4091 FAI images from 2013–2017, that automatically identifies features in the FAI images pixel‐by‐pixel. The identified categories are diffuse aurora, discrete aurora, arc aurora, ambiguous aurora, no aurora, and vignettes. Based on the model output, tags for each image are created and stored in a database, allowing for quick and accurate retrieval of images that contain requested features. Overall, the trained model has a Mean Intersection over Union of 52.8% of all pixels from a validation set generated from 2018–2019 images. Key Points The Fast Auroral Imager (FAI) on CASSIOPE (Swarm‐E) captures auroral images in the near‐infrared (650–1,000 nm) at altitudes from 325 to 1,500 km We train a U‐Net based machine learning model to classify auroral forms pixel‐by‐pixel in the FAI images For unseen data, the model has a Mean Intersection over Union of 53.5% across all the seven class types
 Ionosphere Machine Learning alfven waves auroral imaging magnetosphere M‐I coupling

Details

Metrics

1 Record Views
Logo image