| 英文摘要 | Quasar automatic recognition is a new and interdisciplinary problem on astronomy, pattern recognition, artificial intelligence, signal and image processing. The research on quasar automatic recognition is to assist astrophysicists to explore the mystery of the universe, uncover the features of quasars. Its long-term goal is to build a software package for automatic astronomical spectrum analysis and data processing as required by the LAMOST project. Its short-term objective is to identify quasars from observed astronomical spectrums, to recognize the emission peaks in quasar spectrums, and to determine the red shifts of quasars. To achieve this short-term objective, we thoroughly studied the critical problems in quasar recognition, such as the classification problem of true/false emission peaks in observed spectrums, the matching problem of emission peaks, Hough transform, the learning and training ways in neural networks and the verification problem of red shifts. During this process, we introduced some novel ideas and techniques, proposed and implemented a stratified approach for quasar recognition based on Hough transform and neural networks. The main points of our work are as follows. 1. We collect current knowledge on quasars and emission spectrums, simulate quasar's still emission spectrum and its red shifted emission spectrums. From these emission spectrums, we further clarify the nature of quasar emission spectrums changing with the red shifts. The work provides a sound foundation for further study on quasar recognition. 2. On the basis of the quasar spectrums and astronomical expert knowledge, we propose several ways to recognize emission peaks through analyzing the natures of quasar emission peaks. The ways include emission peaks matching, classifying, back- propagation neural networks and identifying red shifts at first. For each one of the above four ways, we explore and evaluate several specific techniques for its implementation. 3. Due to the low ratio of quasar signals to noise, and lack of expert knowledge, etc. we introduce two approaches to recognize the red shifts, one based on the red shift formula, the other based on Hough transform. Concerning the study of the reliability of red shifts, we propose two verification ways based on still emission spectrum and red shifted emission spectrums respectively. 4. Finally, we present an automatic quasar recognition system by combining artificial neural networks with Hough transform. The system is composed of several functional units, namely preprocessing, peak extraction, true/false peak identification, peaks mapping, Hough transform, red shift determination, final system verification, statistics and learning etc. The system's recognition rate is up to 83% in experiments with 170 observed spectrums. The experiments' results show that the quasar recognition system based on Hough transform has the advantages of simpleness, robustness, efficiency and |
修改评论