|
Title:
|
DEEP LEARNING APPROACH TO CLASSIFYING SOLAR RADIO BURSTS |
|
Author(s):
|
Herman le Roux, Günther Richard Drevin, Roelf Du Toit Strauss and Petrus Johannes Steyn |
|
ISBN:
|
978-989-8704-44-3 |
|
Editors:
|
Hans Weghorn and Pedro Isaias |
|
Year:
|
2022 |
|
Edition:
|
Single |
|
Keywords:
|
Solar Radio Bursts, Space Weather, Deep learning, Convolutional Neural Networks |
|
Type:
|
Full Paper |
|
First Page:
|
13 |
|
Last Page:
|
20 |
|
Language:
|
English |
|
Cover:
|
|
|
Full Contents:
|
click to dowload 
|
|
Paper Abstract:
|
Solar radio bursts have a profound effect on the environment and the technological infrastructure of society. Due to the effects that solar radio bursts have, it is important to be able to correctly classify them in an efficient and timely manner. Recently, there has been great success in object classifying by using convolutional neural networks. Therefore, this paper attempts to classify type II, and type III solar radio bursts, and spectrograms containing no bursts, using a convolutional neural network. An in-depth study was conducted to understand the nature of type II and type III bursts and how convolutional neural networks can be implemented to classify these bursts. Data samples, in the form of FITS files, were collected from the e-Callisto network and processed into images. A convolutional neural network was created with the Python packages TensorFlow and Keras. This model was then trained and tested using the only raw data collected from the e-Callisto network. While the model was not able to reach a very high accuracy the paper shows that such a model can make correct predictions. |
|
|
|
|
|
|
