Speaker
Description
Manual classification of diffuse radio sources is tedious due to the large volume of radio images available from radio sky surveys. Upcoming missions will provide even larger datasets of radio images with diffuse radio emission sources thereby making manual classification impractical due to its time-intensive nature.
We present an alternative method for classification that employs machine learning techniques to categorize diffuse radio sources in radio survey images. We develop a model that uses convolutional neural networks and validate it using a comprehensive dataset of radio cluster images (including synthetic datasets to simulate the diversity expected from future missions).
Provisional results show that the model’s overall classification accuracy is over 87% on the test set, with precision and recall scores exceeding 80% for all classification categories. Our provisional results demonstrate that the proposed framework is not only capable of accurate classification but also scalable to handle the massive influx of data expected from next-generation telescopes, such as the Square Kilometre Array (SKA) and LOFAR 2.0.
| Stream | Science |
|---|