Dim Small Object Detection Technology Based On Spatio-temporal Enhancement

The monitoring and early warning of small celestial bodies plays a vital role in protecting national security and human life safety.When observing these targets,the target energy obtained becomes very weak because the targets are small and far away;At the same time,the observation system is affected by the interference of atmosphere,earth background light,zodiacal light,stars and other backgrounds,as well as various noises in the system,which leads to the low signal-to-noise ratio of the obtained image.Aiming at a series of difficult detection problems,such as small size,weak energy and vulnerable to background noise and clutter,the detection technology of dim small celestial bodies based on spatio-temporal enhancement is studied.The signal and noise of small celestial body imaging system are analyzed,and the signal-to-noise ratio model is constructed.The modeling of dim small celestial body target signal and the statistical characteristics of dim signal are analyzed.Aiming at the problem of dim small object detection with low signal-to-noise ratio,a dim small object detection method based on local region similarity is proposed in this paper.Through the analysis of the characteristics of background photons and the characteristics of certain similarity of background signals in time domain,a new local region similarity factor is proposed to extract moving targets.This method does not directly depend on the gray scale of the image,but extracts the disturbance to the stability of the moving target when it passes through the background to invert the moving target.Therefore,it still shows stable detection performance under low signal-to-noise ratio.In the experiment of real asteroid data,the detection rate is improved by 6.12% compared with other methods.The false alarm rate decreased by 1.32% and the background inhibitory factor increased by0.22.On this basis,a more comprehensive statistical feature space is established by using multiple statistical features including local similarity features,and a dark and weak small object detection method based on statistical feature space extraction and SVM is proposed.The change characteristics of higher dimensions are used to detect the change of target,and the problem of dark and weak small object detection is transformed into the problem of binary classification of target and background,It avoids the threshold segmentation problem that is easy to lead to the increase of false alarm,and has better detection and generalization performance.In the real asteroid data experiment,compared with other methods,the accuracy of positive sample classification is improved by 9.60%.The classification accuracy of negative samples was improved by 5.31%,and the background suppression factor was improved by 0.81.In order to further improve the detection ability of dim targets and adapt to the detection scene with lower signal-to-noise ratio,a spatial enhancement method of dim small celestial bodies based on sub-pixel registration is proposed.Using the method of multi frame accumulation,each frame image is accumulated according to the position of the same spatial target.At the same time,it overcomes the energy diffusion problem caused by the insufficient registration accuracy of the target center in the traditional pixel level registration accumulation method,and can more accurately register the target center position in each frame at the sub-pixel level,so that the target energy is accumulated more intensively and the energy improvement efficiency is higher,so as to track and detect weaker targets.