Radio U-Net: a convolutional neural network to detect diffuse radio sources in galaxy clusters and beyond (2408.10871v1)

Abstract: The forthcoming generation of radio telescope arrays promises significant advancements in sensitivity and resolution, enabling the identification and characterization of many new faint and diffuse radio sources. Conventional manual cataloging methodologies are anticipated to be insufficient to exploit the capabilities of new radio surveys. Radio interferometric images of diffuse sources present a challenge for image segmentation tasks due to noise, artifacts, and embedded radio sources. In response to these challenges, we introduce Radio U-Net, a fully convolutional neural network based on the U-Net architecture. Radio U-Net is designed to detect faint and extended sources in radio surveys, such as radio halos, relics, and cosmic web filaments. Radio U-Net was trained on synthetic radio observations built upon cosmological simulations and then tested on a sample of galaxy clusters, where the detection of cluster diffuse radio sources relied on customized data reduction and visual inspection of LOFAR Two Metre Sky Survey (LoTSS) data. The 83% of clusters exhibiting diffuse radio emission were accurately identified, and the segmentation successfully recovered the morphology of the sources even in low-quality images. In a test sample comprising 246 galaxy clusters, we achieved a 73% accuracy rate in distinguishing between clusters with and without diffuse radio emission. Our results establish the applicability of Radio U-Net to extensive radio survey datasets, probing its efficiency on cutting-edge high-performance computing systems. This approach represents an advancement in optimizing the exploitation of forthcoming large radio surveys for scientific exploration.

• The paper introduces Radio U-Net, a CNN that detects diffuse radio sources with 73% accuracy and 83% recall.
• It employs a tailored U-Net architecture trained on synthetic LOFAR observations to overcome noise and imaging artifacts.
• The network’s success in matching real emission morphology supports its potential for scalable analysis in upcoming large-scale surveys like the SKA.

Radio U-Net: A Convolutional Neural Network to Detect Diffuse Radio Sources in Galaxy Clusters and Beyond

The paper "Radio U-Net: a convolutional neural network to detect diffuse radio sources in galaxy clusters and beyond" by C. Stuardi et al. details the development and application of a fully convolutional neural network, named Radio U-Net, for the detection of diffuse radio sources in radio astronomy. Designed to address the challenges posed by new-generation radio telescope arrays, this work represents a significant contribution to automated image processing in radio interferometry.

Introduction

The forthcoming generation of radio telescope arrays, such as the Square Kilometre Array (SKA), promises substantial advancements in sensitivity and resolution. This progress will enable the identification and characterization of numerous faint and diffuse radio sources. Traditional manual cataloging methods will likely be insufficient to leverage the vast amount of data generated by these new surveys. To bridge this gap, the authors propose Radio U-Net, a convolutional neural network based on the U-Net architecture, tailored to detect extended radio sources such as radio halos, relics, and cosmic web filaments in radio surveys.

Methodology

Radio U-Net Architecture and Training

Radio U-Net's architecture is an adaptation of the original U-Net, tailored to detect diffuse radio sources. The network comprises a contractive path for downsampling and an expansive path for upsampling. It processes images to output a probability map where each pixel represents the likelihood of being part of a diffuse source. The training of Radio U-Net involved synthetic radio observations derived from cosmological simulations, mimicking LOFAR HBA observations, including noise and artifacts typical of interferometric images.

Synthetic Data and Real-world Application

The synthetic data used for training Radio U-Net were based on MHD cosmological simulations incorporating shock-accelerated relativistic electrons emitting synchrotron radiation. These synthetic observations were processed to simulate realistic LOFAR observations, including noise and cleaning artifacts. The primary goal was to train the network to recognize diffuse emission even below the noise threshold and amidst imaging artifacts.

In the real-world application, the trained Radio U-Net was tested on images from the LOFAR Two-metre Sky Survey (LoTSS). The network's performance was evaluated using a sample of galaxy clusters from the PSZ2 catalog. The effectiveness of Radio U-Net was determined by comparing its segmentations with manual classifications.

Results

Radio U-Net demonstrated a significant capability in detecting and segmenting diffuse radio sources. Key results include:

• An accuracy rate of 73% in identifying clusters with diffuse radio emission, with a high recall of 83%, indicating the network's efficiency in capturing most diffuse sources.
• The network showed proficiency in producing segmented images that closely matched the morphology of actual diffuse radio emissions, even in noisy and artifact-laden images.
• Notably, the network's generalization capability extended to identifying extended radio galaxies, highlighting its potential beyond the initial scope of detecting cluster-scale diffuse emission.

Implications and Future Developments

The success of Radio U-Net indicates its practical utility in large radio surveys. The network's ability to automate the detection of diffuse radio sources significantly reduces the computational time and human effort required for data analysis. This is particularly relevant in the context of upcoming large-scale observatories like the SKA, which will produce data at unprecedented scales.

Further development could involve expanding the training set to include a wider variety of simulated scenarios, including radio galaxies and star-forming galaxies, to improve the network's robustness. Additionally, incorporating multi-wavelength data and implementing classification layers could enhance the network's ability to distinguish between different types of diffuse emission, such as radio halos and relics.

Conclusion

Radio U-Net represents a substantial leap forward in the application of deep learning to radio astronomy. By automating the detection and segmentation of diffuse radio sources, it offers a scalable solution to the challenges posed by next-generation radio telescope arrays. Future advancements in network training and architecture will likely enhance its applicability, making it an invaluable tool for astronomers studying the intricate structures of the universe.