Abstract：

With the continuous development and progress of astronomical observation technology, we can explore many galaxies and planets in the distant universe, but most of them require professional equipment. For astronomy enthusiasts and astronomical researchers, it is extremely expensive, which has become a barrier for professionals to explore the starry sky and for the public to understand astronomy. Imaging celestial bodies has become a good solution. Astronomical images can be transmitted at low cost, visualize stars, and are more in line with human visual information. In addition, they provide important data and information for a better understanding of the origin and evolution of the universe. However, there are many challenges in astronomical imaging. For example, the astronomical field is often complex and extremely large in number. How to accurately and clearly image has become a problem to be solved. With the development of computer vision, these difficulties in astronomical image processing have been made possible. This article popularizes the common astronomical knowledge background, summarizes the common astronomical image processing tasks, the difficulties and challenges of the tasks, reviews the processing technology of astronomical images that change over time and the current problems, and looks forward to the future development direction.

Introduction：

Astronomers need to process massive amounts of image data to find and study astronomical phenomena. Therefore, in order to solve the high cost of space exploration equipment, to preserve more celestial information, and to make astronomical research more universal, imaging celestial bodies and processing astronomical images to accurately reflect astronomical information has become the key.

Humans live between heaven and earth and have been exploring the mysteries of the universe since very early times. Therefore, astronomy is one of the oldest sciences. However, the public does not know much about astronomy and it is difficult to acquire astronomical knowledge. Practical difficulties include 1. Serious shortage of astronomical equipment, 2. Difficulty in outdoor observation, and 3. Lack of professional astronomers [1]. However, our agricultural production cannot do without astronomy, and people's daily lives cannot do without astronomy. Breaking down barriers and making information more convenient to spread has become a problem to be solved.

The most important thing is that the main research of astronomical scientists has great practical significance for our lives, such as timekeeping, calendar compilation, and orientation. The development of astronomy has a great impact on human's view of nature. Copernicus' heliocentric theory once liberated natural science from theology.

In summary, a set of convenient and intelligent observation technologies is needed. Imaging celestial bodies has become a good solution, but celestial bodies are complex and huge. How to image and process them has become the content that modern computers need to assist in achieving.

Traditional manual processing methods are time-consuming, labor-intensive and error-prone, and cannot meet the needs of efficient data processing and accurate analysis. The current popular methods of assisting astronomical image processing are: 1. Experienced software processing 2. Crowdsourcing, sharing astronomical images recorded by equipment on a specific application platform, calling on a large number of astronomy enthusiasts to assist in processing 3 Based on computer vision.

I. Main tasks of astronomical images:

1. Galaxy classification and redshift estimation: This is a key task in astronomy, which involves classifying galaxy images into different types (such as elliptical, spiral, etc.) and estimating their redshift values. This is done by identifying and measuring different features in the spectrum.

2. Celestial positioning and tracking: Image processing technology can be used to accurately measure and track the position of celestial bodies, thereby providing important information about the motion and evolution of celestial bodies.

3. Celestial imaging and simulation: Extract celestial features, and generate and analyze celestial simulation data.

4. Abnormal event detection: Detection of ray bursts and pulsars from all-sky radio data.

II. Celestial image processing goals:

1. Removal of noise and artifacts: Celestial images are usually limited by the distance of celestial bodies and a large number of complex factors in the distance space, such as interstellar dust, nearby luminous celestial bodies, and their own instruments. Therefore, the first task of processing celestial images is to remove noise and artifacts.

2. Anti-plane image: Celestial images are collected from the reflecting surface of the telescope and need to be inverted accordingly. This is an extremely simple image processing step that can be completed using a rotation function.

3. Background subtraction: The cosmic microwave background is more obvious in celestial images. It needs to be subtracted to highlight the remaining celestial signals.

4. Image superposition: The celestial images of different bands can be combined into multi-band images to better understand the structure and composition of celestial bodies. However, the spatial range of celestial images is extremely wide and the angles are numerous. Machine vision and other tools are needed to assist in automatic alignment to increase image processing efficiency.

III. Development of astronomical image processing

1.1950s-1970s: This stage mainly relies on hardware and analog technology related to astronomical equipment. Due to the lack of digital computers to process images, astronomers use analog computers and optical turntables to process data. This period is mainly used for the observation and research of a single celestial body or a very limited celestial hemisphere.

2. 1980s-1990s: This period is the digital stage of astronomical image processing technology. Due to the rapid development of computer technology and the widespread use of digital detectors, image processing has developed from hardware equipment to software systems. During digitization, data processing is easier to operate and more efficient than before, and various related software are widely used, such as IRAF (Quincy Analysis Universal Visual Environment), IDL (Interactive Data Language), etc.

3.2000s-2010s: This period is the stage of astronomical imaging, data mining and application. Astronomers can share data through astronomical image processing software. Although the Internet and distributed processing technology can be used to process extremely large astronomical data sets, they are very dependent on manual assistance. The "crowdsourcing" model, in which astronomers publish images on a specific platform and invite a large number of astronomy enthusiasts to jointly process the images, was also a very popular auxiliary processing method at the time. However, due to the technical limitations of the crowdsourcers themselves, they still need to rely on professionals to assist in specifying relevant measures to solve the problem. However, crowdsourcing projects are still an efficient and economical scientific method in the field of astronomy.

4.2020s to present: Computer vision has begun to be widely used in the field of astronomical image processing, and can process large-scale, multi-dimensional, and complex astronomical data sets, greatly improving processing efficiency.

Application of computer vision in astronomy

Machine learning is a method of implementing artificial intelligence, mainly used for problems that are difficult to describe with rules and explicitly program. The goal is to study how to make computers simulate human learning behavior, automatically improve algorithms through experience, learn implicit patterns from data and build models, so as to make predictions for similar problems. Machine learning methods are widely used in many natural science fields such as medicine, biology, physics, astronomy, etc., providing these disciplines with new ideas for solving problems in the era of big data.