Research On Automatic Search And Recognition Technology Of Space Targets In High Orbits

With the increase of human spaceflight activities,the number of spacecrafts working in earth orbits has increased significantly,which inevitably generates hundreds of millions of space debris,posing a serious threat to the satellites and spacecrafts.It is important to accurately identify,locate and track space targets in earth orbits,which can predict possible collisions and then control satellites to avoid potential threats.In addition,space target surveillance is also an important guarantee for national security,so the research on space target detection technology is of great significance and value.Space targets in high orbits are far from the ground,and have the characteristics of little scales and dim brightness.Under the cover of a large number of stars,there are many technical difficulties in realizing automatic search and recognition of space targets with small sizes.First of all,star image processing and target recognition in complex backgrounds still have big limitations,especially when the bright target,cloud,optical vigniting,detector uniformity and other factors exist,making the recognition and tracking ability greatly weakened.At the same time,the target detection mode of first recognizing and then tracking is easily interfered by stars and false targets,which makes the target positioning accuracy low and the error of target trajectory prediction large.Moreover,current observation equipments and processing technology are often non-intelligent and cannot provide real-time feedback to the telescope according to image data,making the effect of shooting and tracking nonoptimal and making the back-end processing work a great challenge.This paper uses the telescopes with a large diameter and a wide field for observing conditions,by studying the characteristics of space targets,designing image processing algorithms and developing targeted observation ways,to achieve automatic search,detection and measurement for multitargets with little scales in the images with a large field of view,to implement an intelligent decision-making system that automatically adjusts tracking modes of telescopes based on image feedback and to provide more accurate data for space early warning,cataloguing and orbit determination.The specific work of this paper is as follows:1.Under the influence of moonlight,clouds and optical vignetting of the system,it is difficult to effectively identify and segment star images in complex backgrounds only by relying on gray threshold methods.Aiming at the problem,we study the methods of image background modeling and gray-level threshold segmentation,analyze the grayscale characteristics of complex background images,and combine imaging features of the optical system,to propose a star image segmentation method based on correlation between space targets and one-dimensional Gaussian form.The method shifts the processing core from the traditional gray threshold to correlation coefficient threshold,and at the same time,the standard deviation of local data and the ratio of mean square error between the target and the model to variance of local data are used to remove false alarms.The results are compared and analyzed with different gray level threshold segmentation methods for verifying the validity of our method.The results show that for real star images in complex scenes,our proposed method can identify more targets,and that for two groups of simulation star images in complex backgrounds with 2000 frames each,the recognition rates and false alarm rates realized by our proposed method are respectively 97.6%,0.43% and 97.9%,0.48%.2.In traditional space target detection processes,interframe registration should be carried out first.After detecting targets,the morphological characteristics should be analyzed,and then motion parameters should be calculated.Under the influence of bright stars,there are some problems such as inaccurate target recognition and positioning,large deviation of target displacement calculation,etc.For solving the problems,the paper analyzes the motion characteristics of the targets and image backgrounds from the perspective of time domain and space domain,and proposes a reverse procedure detection method of space target streaks based on motion parameter estimation.The target displacement is detected first through the phase differences between two image frames,and then the target is searched for and located,which is completely opposite to the traditional detection procedure and effectively inhibits the interference from bright stars.The results show that our proposed method can estimate the displacement,length and position of the target streak with sub-pixel accuracy for real star images of different scenes.For five groups of simulation images with 100 frames each,the Gaussian noise with the standard of 0,5 and 10 is added respectively,and the following estimation accuracies of target streak parameters are achieved: estimation accuracies of target displacement vectors are 0.08,0.31 and 0.48pixels;estimation accuracies of target lengths are 0.04,0.14 and 0.26 pixels;estimation accuracies of streak angles are 0.07°,0.3° and 0.5°;target positioning accuracies are at subpixel level.3.When there are multiple targets with different motion states in the field of view,the traditional target detection methods need to analyze all morphological parameters of the targets and to carry out interframe matching,and then calculate the motion characteristics and track them.However,when the detection algorithm is unable to accurately analyze the shapes of each target,the multitargets cannot be matched,so that more frames need to be taken for analysis.To solve this problem,the reverse procedure detection method of multiple targets in star image sequence is proposed,which extracts the displacement vectors of multiple targets from phase difference spectra,and then directly catalogs,tracks and locates the multiple targets,overcoming the problems in the procedure of removing false targets.For the simulation star image sequence with five different shapes of target streaks,the Gaussian noise with the standard of 0,5 and 10 is added respectively,and the following estimation accuracies of target streak parameters are achieved: estimation accuracies of target displacement vectors are 0.06,0.32 and 0.55 pixels;estimation accuracies of target lengths are 0.05,0.21 and 0.41 pixels;estimation accuracies of streak angles are 0.16°,0.42° and 0.7°;target positioning accuracies are at subpixel level.