Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

PI-AstroDeconv: A Physics-Informed Unsupervised Learning Method for Astronomical Image Deconvolution

Shulei Ni, Yisheng Qiu, Yunchun Chen, Zihao Song, Hao Chen, Xuejian Jiang, Huaxi Chen

TL;DR

This work proposes an unsupervised network architecture that incorporates prior physical information and adopts an encoder-decoder structure while leveraging the telescope's PSF as prior knowledge to achieve image deconvolution and reconstruction.

Abstract

In the imaging process of an astronomical telescope, the deconvolution of its beam or Point Spread Function (PSF) is a crucial task. However, deconvolution presents a classical and challenging inverse computation problem. In scenarios where the beam or PSF is complex or inaccurately measured, such as in interferometric arrays and certain radio telescopes, the resultant blurry images are often challenging to interpret visually or analyze using traditional physical detection methods. We argue that traditional methods frequently lack specific prior knowledge, thereby leading to suboptimal performance. To address this issue and achieve image deconvolution and reconstruction, we propose an unsupervised network architecture that incorporates prior physical information. The network adopts an encoder-decoder structure while leveraging the telescope's PSF as prior knowledge. During network training, we introduced accelerated Fast Fourier Transform (FFT) convolution to enable efficient processing of high-resolution input images and PSFs. We explored various classic regression networks, including autoencoder (AE) and U-Net, and conducted a comprehensive performance evaluation through comparative analysis.

PI-AstroDeconv: A Physics-Informed Unsupervised Learning Method for Astronomical Image Deconvolution

TL;DR

This work proposes an unsupervised network architecture that incorporates prior physical information and adopts an encoder-decoder structure while leveraging the telescope's PSF as prior knowledge to achieve image deconvolution and reconstruction.

Abstract

In the imaging process of an astronomical telescope, the deconvolution of its beam or Point Spread Function (PSF) is a crucial task. However, deconvolution presents a classical and challenging inverse computation problem. In scenarios where the beam or PSF is complex or inaccurately measured, such as in interferometric arrays and certain radio telescopes, the resultant blurry images are often challenging to interpret visually or analyze using traditional physical detection methods. We argue that traditional methods frequently lack specific prior knowledge, thereby leading to suboptimal performance. To address this issue and achieve image deconvolution and reconstruction, we propose an unsupervised network architecture that incorporates prior physical information. The network adopts an encoder-decoder structure while leveraging the telescope's PSF as prior knowledge. During network training, we introduced accelerated Fast Fourier Transform (FFT) convolution to enable efficient processing of high-resolution input images and PSFs. We explored various classic regression networks, including autoencoder (AE) and U-Net, and conducted a comprehensive performance evaluation through comparative analysis.
Paper Structure (12 sections, 3 equations, 3 figures, 1 table, 1 algorithm)

This paper contains 12 sections, 3 equations, 3 figures, 1 table, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Works
  3. Our Method
  4. Overview
  5. Physics-informed Network
  6. Accelerate convolution through Fourier Transform
  7. Loss Functions
  8. Experiments
  9. Datasets and Experimental Setting
  10. Results and Discussion
  11. Conclusion
  12. Acknowledgments

Figures (3)

  • Figure 1: The PI-AstroDeconv architecture incorporates U-Net as its backbone. The left side of the U-Net represents the input, with the right side incorporating the Point Spread Function (PSF) or telescope beam. According to our network architecture design, during the training phase, the input and output of the network are expected to remain consistent, irrespective of the choice of backbone network. After the training is completed, during the inference phase, the backbone network will directly generate the inference results, which are represented as predictions at the bottom of the diagram.
  • Figure 2: Sketch of telescope observation effects. The image on the left corresponds to the clean image, and the image in the middle showcases the PSF or beam of the telescope. The image on the right illustrates the blurring effect brought by the PSF or beam. The symbol in the middle represents the convolution operation.
  • Figure 3: The input/output, and deconvolved images of the PI-AstroDeconv network architecture. The left panel illustrates both the input and output of the PI-AstroDeconv network architecture. In the middle, the image displays the deconvolutional result obtained using the Autoencoder. The right panel showcases the deconvolutional outcomes achieved with UNet. The horizontal and vertical coordinates in the figure represent right ascension and declination, respectively, indicating the corresponding position on the celestial map.