Real-time Astronomical Target Detection Method For Wide Field Small Aperture Telescope

The Wide-field Small Aperture Telescope(WFSAT)is an important optical instrument for astronomical observation.Due to it has a large field of view and low cost,WFSAT is suitable for continuous sky surveying at different places,and it has become a key instrument for time-domain astronomical observations.Since time-domain astronomy is interested in observation of moving targets or targets with variable magnitudes,WFSATs mostly obtain data by short-exposure.After obtaining these data,astronomical targets need to be extracted from the observation data immediately for discovery of key targets such as supernovae and gravitational wave electromagnetic counterparts.Nowadays,SExtractor and other source extraction algorithms are commonly used in astronomical target extraction.But these algorithms need to adjust parameters to ensure the accuracy rate of extracting different astronomical targets from different images.At the same time,such algorithms have some problems,such as low efficiency in detection of dark sources,low efficiency in detection of targets near bright sources and they can only detect targets without classifying them.To solve problems mentioned above,this paper proposes a deep learning based real-time astronomical target extraction method.Considering WFSATs are usually arranged as telescope arrays in the observatory.Even with the same parameters and different WFSATs point to the same sky area for data acquisition,observed targets may have different shapes,which would make the deep learning based astronomical target extraction method with unstable target extraction results when it is good or bad.However,on the other hand,integrating target extraction results of these WFSATs,it is possible to obtain better target extraction results than that of a single WFSAT.Thus,based on the real-time astronomical target extraction method for single WFSAT,this thesis further propose a target extraction method for WFSAT arrays that would be beneficial to improve the efficiency of target processing and the observation ability of WFSAT arrays,which would be the key data processing technology for WFSAT arrays.In this paper,a deep learning based astronomical target extraction algorithm is proposed,the results show that our algorithm has better performance than that of traditional methods.The object extraction algorithm is further proposed to be deployed in an embedded device with a WFSAT to form an observation unit.The observation unit can obtain positions and types of astronomical targets and broadcast extraction results in the observation net with real time.With the detection results from observation units,an ensemble learning algorithm is proposed,which can be used to process detection results from WFSATs.It has been proved that the astronomical target extraction method for WFSAT arrays can improve the observation ability of faint targets.The algorithm can provide new data processing methods for WFSATs.