Accueil Imprimer Annuaire Plan du site Crédits Fil RSS du site Twitter Plans d'accès Contacts Annuaire Webmail Intranet Logo

Accueil > Plus d’actualités > An application of Deep Learning to the automatic detection of Interplanetary Coronal Mass Ejections

An application of Deep Learning to the automatic detection of Interplanetary Coronal Mass Ejections

ICMEs are the interplanetary counterpart of Coronal Mass Ejections, the expulsion of large quantities of plasma and magnetic field in the interplanetary medium. They are generally identified by characteristics such an enhanced and smoothly rotating component, low proton temperature and declining velocity profile. As these events are measured when they come across the trajectory of a spacecraft measuring the solar wind physical parameters (WIND, ACE, STEREO, …), their detection is affected by a strong variability and a difference of interpretability from an user to another. Which makes their detection time consuming and fastidious. Consequently, the existing ICMEs catalogs are non-exhaustive, subjective and hardly reproducible. Which constitute a bottleneck of the statistical studies performed on these phenomena.

Using convolutional neural networks, a class of deep learning algorithms especially used to detect objects in images , on the 20 years of data provided by the spacecraft WIND, Nguyen et al. (2019) established an automatic, fast, reproductible and adaptable detection method of ICMEs.
The predictions made by this method on the globality of the data provided by WIND can be accessed here : https://hephaistos.lpp.polytechnique.fr/data/machine_learning/ICME/index.html

JPEG - 315.4 ko
Vue d’artiste d’une Ejection de Masse Coronale Interplanétaire (Encadré orange), prédiction de la méthode sur 3 mois de donnée (première figure) et prédiction de la méthode sur un seul évenement. Les couleurs du bleu vers le rouge indiquent une probabilité croissante de détecter un événement.

In addition to detecting a great number of ICMEs while making few errors, the method also offers a unambiguous visual proxy of ICMEs useful for an external data observer.
The generality of the method, that requires no physical knowledge specific to ICMEs, paves the way for the automatic detection and the massive statistical studies of many different event signatures in spacecraft measurements.

Voir en ligne : https://iopscience.iop.org/article/...


transparent
Tutelles : CNRS Ecole Polytechnique Sorbonne Université Université Paris Sud Observatoire de Paris Convention : CEA
transparent
©2009-2019 Laboratoire de Physique des Plasmas (LPP)

Mentions légales
Exploitant du site : Laboratoire de Physique des Plasmas, Ecole Polytechnique route de Saclay F-91128 PALAISEAU CEDEX
Hébergeur : Laboratoire de Physique des Plasmas, Ecole Polytechnique route de Saclay F-91128 PALAISEAU CEDEX
Directeur de la publication : Pascal Chabert (Directeur)