| Summary: | In this work we have developed an automated method for detecting Spread-F in
ionograms based on machine learning (ML) algorithms. Spread-F is a feature
that appears on an ionogram when there are specific ionospheric irregularities
at the time the ionogram is taken. Our work contributes in three key ways: 1)
help other researchers quickly decide on the most suitable Spread F detection
methodology, 2) provide publicly available labelled ionogram dataset for others
to use to build their own SF classification models and 3) develop an interactive
web application through which one can view ionograms, check whether a given
ionogram has SF, as well as vote on already classified ionograms. The methods
explored in this work include Support Vector Machines and convolutional neural
networks (CNN). For CNN, we created models based on existing public neural
network architectures in a process called transfer learning. Transfer learning
outperformed the other approaches and the best performing architecture was
the one based on ResNet50.
|
|---|