CNN-based detection and classification of UV-radiation solar and extrasolar flares
This MSc. thesis studies the applicability of the CNN-based algorithm stella [2] for the detection of flares from different data sets: optical TESS (exo)stellar, EUV DSO/AIA and GNSS (the last two for Solar flares). First, the optical TESS (exo)stellar part, makes use of the new TESS catalog of flar...
| Autor: | |
|---|---|
| Tipo de recurso: | tesis de maestría |
| Fecha de publicación: | 2022 |
| País: | España |
| Institución: | Universitat Politècnica de Catalunya (UPC) |
| Repositorio: | UPCommons. Portal del coneixement obert de la UPC |
| Idioma: | inglés |
| OAI Identifier: | oai:upcommons.upc.edu:2117/386398 |
| Acceso en línea: | https://hdl.handle.net/2117/386398 |
| Access Level: | acceso abierto |
| Palabra clave: | Solar flares Stellar activity Global Positioning System Neural networks (Computer science) CNN neural networks flares solar flares stella stellar flares GNSS solar demon TESS sdo/aia flare detection Erupcions solars Activitat estelar Sistema de posicionament global Xarxes neuronals (Informàtica) Àrees temàtiques de la UPC::Física::Astronomia i astrofísica |
| Sumario: | This MSc. thesis studies the applicability of the CNN-based algorithm stella [2] for the detection of flares from different data sets: optical TESS (exo)stellar, EUV DSO/AIA and GNSS (the last two for Solar flares). First, the optical TESS (exo)stellar part, makes use of the new TESS catalog of flares. It is capable of training new models on uni-sector data, thus avoiding the over-fitting present, indicatively, in the pre-training models of stella. An analysis of the behavior of the models in sectors other than those where they have been trained and according to the spectral type of the flaring stars is presented. AUC across 2 × 10 (exo)stellar models trained span [0.973, 0.981]. Second, we consider the EUV SDO/AIA and GNSS data sets for Solar flares detection. With these two data sources, stella models are created for the detection of solar flares. Due to the casuistry related to the scarcity of data available in these two data sources used (due to low number of cadences for SDO/AIA or due to short baseline for GNSS), the results are not entirely satisfactory. However, they point the improvement paths for future development of better flare detection models based on stella. |
|---|