Deep learning for morphological identification of extended radio galaxies using weak labels

Abstract

The present work discusses the use of a weakly-supervised deep learning algorithm that reduces the cost of labelling pixel-level masks for complex radio galaxies with multiple components. The algorithm is trained on weak class-level labels of radio galaxies to get class activation maps (CAMs). The CAMs are further refined using an inter-pixel relations network (IRNet) to get instance segmentation masks over radio galaxies and the positions of their infrared hosts. We use data from the Australian Square Kilometre Array Pathfinder (ASKAP) telescope, specifically the Evolutionary Map of the Universe (EMU) Pilot Survey, which covered a sky area of 270 square degrees with an RMS sensitivity of 25–35 $\mu$ Jy beam $^{-1}$ . We demonstrate that weakly-supervised deep learning algorithms can achieve high accuracy in predicting pixel-level information, including masks for the extended radio emission encapsulating all galaxy components and the positions of the infrared host galaxies. We evaluate the performance of our method using mean Average Precision (mAP) across multiple classes at a standard intersection over union (IoU) threshold of 0.5. We show that the model achieves a mAP $_{50}$ of 67.5% and 76.8% for radio masks and infrared host positions, respectively. The network architecture can be found at the following link: https://github.com/Nikhel1/Gal-CAM

RG-CAT: Detection pipeline and catalogue of radio galaxies in the EMU pilot survey

“…We create cutouts of the same size as the radio images and then reproject the infrared images onto the radio images using the world coordinate system. In contrast to radio images, infrared images undergo noise reduction processing using the method detailed in Gupta et al (2023Gupta et al ( , 2024. Examples of these processed infrared images are showcased in the right columns of Fig.…”

“…In weakly supervised learning, indirect labels are leveraged for the entire training dataset, serving as a supervisory signal. This specific approach has found utility in the classification and detection of extended radio galaxies (Gupta et al 2023). In supervised learning, the model undergoes training using image-label pairs, where these labels provide complete information required for the model to make specific predictions.…”

“…(2023), and Gupta et al. (2024), have found application in the morphological classification and detection of radio galaxies.…”

“…2023a) designed to address the challenge of associating radio galaxy components (Gupta et al. 2024). The RadioGalaxyNET dataset was curated by professional astronomers through multiple visual inspections and includes multimodal images of the radio and infrared sky, along with pixel-level labels on associated components of extended radio galaxies and their infrared hosts.…”

“…Using this comprehensive dataset of extended and compact radio galaxies and their infrared hosts, we train the Gal-DINO multimodal model (Gupta et al. 2024) to simultaneously predict bounding boxes for radio galaxies and potential keypoint positions of their infrared hosts, where a keypoint in ML refers to a specific point or landmark in images. These detections are subsequently employed to generate the first catalogue of compact and extended radio galaxies observed in the first EMU pilot survey (EMU-PS) conducted with the ASKAP telescope.…”