Astronomical Image Screening And Study Based On Support Vector Machine

At present,time domain astronomy is a very active research field.Its main research objects include the target source of continuous changes in luminosity such as supernovae,variable stars and extrasolar galaxies and the fast moving targets such as near earth objects in the solar system.In astronomical research in time domain,automatic survey of small and medium-sized optical telescopes is an important source for acquiring observation data.During observation,the telescope automatically observes and exposes continuously according to the observation strategy to acquire large amounts of astronomical images densely sampled on time.Influenced by this model,a large amount of data obtained from the external environment,telescopes,and CCD will produce some defects.These defect data will be transmitted along with effective data which greatly wastes storage and network resources,and affects processing and scientific research of later data simultaneously.For this reason,it is very important for data transmission and automatic processing to automatically identify and filter the defective data when obtaining data.The CCD is an important source of image noise.In order to better studycharacteristics of the image noise,the imaging process of CCD/CMOS sensor is simulated,the process from photons to electrons,from electrons to analog voltages and from analog voltages to digital voltages is performed,a high precision photoelectric sensor model is constructed,and the influence of image aberration on image quality is simulated.Besides,several kinds of astronomical image noise are simulated starting from several common factors affecting the quality of astronomical images: point defect noise,bar like defect noise and cloud noise etc.According to several major factors influencing the image data of the outside world,the method that the following four kinds of image evaluation indexes are fused by using support vector machine is proposed in this paper: grayscale inconsistency,point defect,bar like deficiency,cloud noise,etc in order to realize the different quality of astronomical image filtering and use the astronomical image of GPU technology to achieve the rapid screening of requirements according to the observation data processing speed requirements of targets such as transient source.After screening the defective image data,the cloud index graph is extracted for the data with cloud noise.The experimental results show that the classification accuracy rate of astronomical images through the system can reach more than 97% and the speed can reach 3.8s per image of image screening which meets the needs of rapid screening.Using the histogram transform and feature extraction can extract index map of the cloud preliminarily which provides a reference to the light anddim target detection.