Automatic Recognition Of Space Targets In Complex Background

With the continuous expansion of the scale of the development and utilization of space technology,space has become an important platform for people to obtain and transmit information,so the control of the space appears particularly important.Space control mainly refers to the monitoring of space targets,including the detection,recognition and tracking of space objects.Space targets include satellites and space debris,and all kinds of universe flying objects entering the earth in outer space.The accurate identification and location of space targets is the main purpose of the space target surveillance,and an important part of space situation awareness,and also an important technical support to ensure the safety of manned space flight.Therefore,it has important practical significance and application value to study the space target detection technology.In this paper,the ground-based electro-optical detection system is used to detect and track the space targets.The advantages of this detection method is high accuracy,strong intuitive,low investment cost,not affected by ground clutter,but low efficiency,great affected by weather changes.In order to overcome these shortcomings,we use large field optical telescope to search and measure the space targets in the distance,so as to achieve rapid identification of space targets.But the large field optical system itself has the optical distortion,uneven background,high requirement of real-time image processing,and the existence problems of space targets in distance,low brightness,too many background stars,so how to automatically identify multiple,dark,weak targets in the condition of complex background is the main purpose of this thesis.This paper research how to enhance the ability of space targets recognition and tracking and improve the accuracy of extraction of space target centroid according to the study of the space target measurement photoelectric telescope system ?astronomical image processing?space target detection and recognition and space target centroid extraction.The completed research work as follows:(1)this paper studies the space target photoelectric telescope system,including optical system?mechanical structure and CCD detector.(2)this paper studies the image processing techniques,including image denoising,image enhancement techniques and image threshold segmentation to improve the image signal to noise ratio.According to the analysis of the astronomical image noise,a new adaptive extremum median filtering denoising algorithm is put forward based on energy function,which employs a twice-check strategy to reduce the false detection ratio of noisy pixels and uses the improved adaptive median filter and the energy function model to recovery noise imagery.The simulated and real star map experiments show that,the Peak Signal to Noise Ratio(PSNR)is improved about 3 times and the Mean Squared Error(MSE)is reduced by 3.16×10-5 in the terms of objective evaluations,the proposed method can effectively improve the denoising result and thus is applicable to star maps.(3)This paper analysis the motion characteristics of space targets.According to the difference in motion between the space targets and the background of the stars,the “moving point object detection algorithm from faint space based on temporal-spatial domain” is put forward,making full use of the distribution information of space targets in the time domain,combining with the spatial correlation.Target motion trajectory is finally extracted based on correlation coefficient matrix statistical information,and the velocity estimation model of the moving target is built.This paper also proposes an evaluation method,which combines detection probability and false alarm probability,to verify this method.The experimental results demonstrate that the proposed method outperforms the compared methods and can achieve high detection probability while keeping low false alarm probability.Compared with simply expanding telescope diameter,this method provides a higher performance-price ratio way to improve the ability of space target detection.The traditional moving target detection techniques for star maps are sensitive to factors such as frame brightness,frame registration and imaging mode.To address this issue,a new moving target detection method based on distance matrix is put forward,which takes advantage of the fact that the topological structure of targets is stable.The new method detects targets in each frame individually and then constructs the distance matrix of targets,then calculates the difference of distance matrices of adjacent frames.The distance change for moving targets is much greater than that of the stationary targets,which can be used to separate the moving targets from stationary targets.Since the distance matrix is independent of imaging condition for star maps and is only affected by moving targets,the algorithm is robust to frame brightness change,frame nonregistration and imaging mode.Experiments for simulated and real data demonstrate that,the method can detect moving targets from background stars effectively and has lower false alarm ratio,even if the frames are unregistered.This paper presents a new space moving target detection method based on time domain features.We construct the time spectral data,then analyze the time domain features of the main objects(targets,stars and the background)in star maps,finally detect the moving targets using single pulse feature of the time domain signal.The experiment of real star target detection shows that the proposed method can effectively detect the moving targets trajectory in the star map sequence,and achieves 99% detection probability when the false alarm rate is about 8×10-5,which outperforms the compared algorithms.(4)According to the shortcomings of the traditional centroid extraction algorithm,the paper puts forward the anisotropic Gaussian surface fitting model,the model by using two different Gaussian blur parameters and rotation factors to capture target of anisotropic characteristics of different directions,is suitable for the fuzzy random direction caused by satellite movement.Simulation experiments and real data test show that the overall positioning accuracy the method of can be achieved respectively 0.008 and 0.04,able to accurately extract the map target centroid,and has improved greatly than the traditional methods.