Research On Automated Processing And Target Detection Of Astronomical Information

It is a permanent topic for human beings how the universe is formed, developed and evoluted. The universe contains all of space, time, material and energy. The research level of universe marks the position of a country in the frontier of science and technology and has special significance to many disciplines, being the great impetus to modern science and technology, especially to the sophisticated space technology development.There are two basic ways for modern cosmic research in visible bands:stellar spectrum and astronomical image. The former is one-dimensional data, while the latter is two-dimensional data. They have close relationship between them. By analyzing spectra, people can measure many chemical components and physical parameters of stellar objects, qualitatively or quantitatively, such as the chemical composition, the surface temperature, the luminosity, the diameter, the quality, the radial velocity and rotation, directly or indirectly. Therefore, spectral analysis occupies an important position in astronomy and astrophysics. Astronomical image analysis provides another way to research celestial bodies. Using astronomical image, we can study the target morphological structure of celestial body, including celestial age, state, evolution tendency, and provide important technical support for astronomical observation and study.The National Major Scientific Project LAMOST (Large Sky Area Multi-Object Fiber Spectroscopic Telescope) was set up in 1997 and completed an adoption of national acceptance in June,2009 and was put into operation since then. The aperture of LAMOST is 4m, enabling it to obtain the spectra of objects as faint as down to 20.5 magnitudes. Within a 5°field of view, it may accommodate as many as 4000 optical fibers. So the light from 4000 celestial objects will be led into a number of spectrographs simultaneously. Thus the telescope will be the one that possesses the highest spectrum acquiring rate in the world. The observation targets of LAMOST are galaxies and stars in the universe. With the operation of LAMOST, thousands of spectra at every observation night can be obtained and the total volume of spectrum data can as many as billions of bytes. So the automated identification and analysis of spectra data is a challenging task. It needs the methods of automated identification and analysis, including spectra identification, classification and parameter measurements.WOSDU (Weihai Observatory of Shandong University) is set up to find asteroid and supernova. Every day, it generates large volume of astronomical image data. In asteroid detection, for example, many celestial images are produced everyday. The size of each image is 2048×2048×16bit,8M. To find new asteroids, we must first detect the celestial bodies in the image, and then transform the pixel coordinates to celestial sphere coordinates. After that, a matching process is carried out in the celestial catalog to seek out whether the targets are already in the catalog or not. If not, the targets will be considered as asteroid candidates and observed later, and their orbits will be calculated for further identification. Because of the significant number of images, a lot of processing must be involved. The target detection is one of the important processing steps. In the target detection, the accuracy and speed are the most important factors to be considered. Our effort is to find a method to detect the targets in the image fast and effectively. It follows that the fast and effective processing for such huge amounts of astronomical data puts forward an urgent need to computer information technology. In the meantime, the mass characteristics and openness of astronomical data open a new research and application field for computer information technology. The timely and effective processing of such massive data needs to resort to many modern information processing technologies, such as image processing, data mining, and signal processing, and so on.This project, sponsored by National Science Foundation Fund, focuses on the key problems in astronomical research field, such as the astronomical data pretreatment, target detection from astronomical image, the classification of high-dimensional data, the data mining of rare celestial objects and so on, by means of data mining, image processing, artificial intelligence, signal processing and other advanced information technologies. The astronomic data to be used are from LAMOST and WOSDU. A series of effective algorithms are designed, verified and will be put into use for LAMOST celestial spectral automatic identification and analysis system. The study is an interdisciplinary research between astronomy and information; it is a concrete application of the latest computer information technologies to in the field of astronomy, so as to achieve new scientific achievements. Therefore, this research has a very good theoretical research and practical application value.The key content of this project includes four aspects:astronomical data pretreatment (spectra denoising) and emission line identification; target detection in astronomical image, celestial spectra classification and data mining of rare celestial bodies.The following research and innovation works have been done respecting to the key issues mentioned above:Firstly, our research focuses on the astronomical spectrum denoising. One-dimensional astronomical spectra observed by the astronomical spectroscopic telescope are often degraded by noise in the acquisition phase and have a poor signal-to-noise ratio. The reduction of noise is highly desirable for a lot of reasons. Noise reduction, as an integral part of signal estimation, has been an extensively studied for many years in the signal processing community. The goal of signal denoising is to recover the original signal from noisy signal corrupted by additive noise. Over the past twenty years, there has been considerable interest in the use of wavelet transforms for removing the noise from signals. Many thresholding rules based on the orthonormal wavelets have been proposed. However, as Coifman and Donoho pointed out, the denoising algorithm based on the orthonormal wavelets exhibits pseudo Gibbs phenomena in the neighborhood of discontinuities. Therefore, they proposed a translation-invariant denoising scheme to reduce the artifacts. In addition, it has been shown that a redundant representation is substantially superior to a nonredundant representation for signal denoising in term of mean-squared error and signal-to-noise ratio. Therefore, the translation-invariant redundant transforms are desirable for spectra denoising. The dual-tree complex wavelet transform introduced by Kingsbury is redundant and near translation-invariant. In this paper, we propose a spectra denoising algorithm based on the dual-tree complex wavelet transform (DTCWT), which can give higher signal-to-noise ratio and better quality. In astronomical spectra preprocessing, the adaptive denoising method has been studied based on the dual-tree complex wavelet transform. This method adjusts the complex wavelet coefficients by MAP estimation theory in order to suppress the noise. On the base of protecting the spectral line and other important information, this method suppresses the noise and pseudo Gibbs phenomenon, improves the efficiency of the algorithm, and provides an efficient tool for later spectra processing. In addition, we proposed a new method to identify emission line stars (ELS) spectra automatically. Stellar spectra are characterized by obvious absorption lines or absorption bands, while those with emission lines are usually special stars such as Cataclysmic Variable stars (CVs), HerbigAe/Be etc. The further study of this kind of spectra is meaningful. The new method to identify emission line stars (ELS) spectra is like this:after the continuum normalization is done for the original spectral flux, line detection is made by comparing the normalized flux with the mean and standard deviation of the flux in its neighbor region. The results of the experiment on massive spectra from SDSS DR8 indicate that the method can identify ELS spectra completely and accurately. Since no complicated transformation and computation are involved in this method, the identifying process is fast and it is ideal for the ELS detection in large sky survey projects like LAMOST and SDSS.Secondly, our research is on the stellar object detection. In astronomical research, the Charge Coupled Device (CCD) images obtained from celestial observations are usually stored in Flexible Image Transport System (FITS) format and the sizes of such images are usually very large,8 Megabytes or more being not unusual. In addition, the amount of such images is great due to continuous celestial observations. So the demand of real time processing of these large images is challenging. This paper proposes an efficient method to detect objects, which is an important step in astronomical image processing, by designing a scan accelerator and a recursive measure routine. A method to detect target in astronomical image has been designed and implemented based on space domain. The experiments showed that by scanning in a recursive way and using a designed accelerator, the detection process can be greatly sped up and multi-objects can be detected quickly and many other parameters of the targets can be obtained. In addition, a 3-D visual system has been established. The stars that met the specific criteria in a certain sky region can be retrieved from UCAC2 catalog and their distribution can be displayed in the 3-D celestial sphere system. In this way, the retrieval information can be visualized.Thirdly, a scheme to optimize the Random Forests classification parameters for astronomical spectra using Genetic Algorithm has been proposed. As an efficient and stable algorithm for high-dimensional spectral data classification, random forest has some advantages compared to other algorithms in the efficiency and accuracy. Random forest’s computation efficiency and classification accuracy are affected by the number of trees and the number of attributes randomly selected. The appropriate thresholds are chosen to minimize training time and get higher accuracy while ensuring that the accuracy is within an acceptable range. Appropriate threshold of the number of trees will obtain the shortest training time while ensuring the accuracy. In addition, the appropriate number of random attributes will make the training time minimized. The threshold is relevant to the data. Therefore the threshold selected will affect efficiency and accuracy. In this paper, we proposed a scheme which can quickly determine the parameters needed by Random Forests in classification applications and improve the automation and intelligence of classification by avoiding the manual parameter estimation. This method can also improve the classification accuracy and reduce the training time of the classifier.Finally, a data mining method for searching cataclysmic variables has been proposed. This method can screen out rare celestial bodies effectively by combining PCA and BP ANN. By PCA dimension reduction, the high dimensional data space has been greatly reduced, and after that, BP ANN is used to screen out rare celestial objects with high accuracy. In this way, the training time becomes shorter. Experiments show that the proposed method is effective for the detection of specific celestial objects. The method can be used not only to search the Cataclysmic Variable stars but also to find other types of special objects. It can greatly reduce the power of work and time of manual processing. Due to the high speed, it can mainly meet the LAMOST quasi-real-time spectral processing requirements. With a parallel data processing environment, the data input, dimension reduction, data mining and other operations can be done simultaneously, thus improving the outcome of scientific research.