Research On Key Technologies Of Detectiong And Recognition Via Space-based Optical Surveillance Systems

The space-based optical surveillance system is called “space sentinels” against the threat from space,because of its important advantages,such as wide-area,24-hour service,over national boundaries surveillance.However,the structure and function of space-based optical surveillance system have been changing.The new changes make the detection and recognition in information processing system a challenging task.This paper discusses the detecting and recognizing technology within the context of the space-based optical system against the aerospace threat,and focuses on track-before-detect of dim small targets,detection and tracking of small extended targets,the optical passive ranging for boost-phase targets under single satellite and the midcourse target discrimination.The main work and research achievements are as follows:In Chapter 2,the track-before-detect method of dim small targets is studied.For dim small infrared targets,they often submerge in complex clutter background.Track-before-detect method is considered as an effective solution.The labeled multi-bernoulli(LMB)filter is applied to track-before-detect(TBD)of infrared small targets.The LMB TBD algorithm is proposed for linear small targets in this work.In order to detect and track multiple maneuvering small targets with time-varying motion models,an IMM-LMB TBD algorithm is further proposed by integrating the interacting multiple models(IMM)with LMB TBD.The proposed algorithms can not only inherit the merit of LMB filter such as true trajectory filtering,but also more accurately track multiple small targets under the low signal-to-noise ratio(SNR).In Chapter 3,the detecting and tracking of size-varying targets are studied.The size-varying target is considered as one subclass of extended target.Its shape is spot and similar to classical point target.Howerver,the target size varies in size space.For detection,the target size is firstly seleted,and then target is identified.To do this,the multiscale minimum target intensity filter is introduced to find the optimal target size.The complete detection is achieved by integrating morphology low-pass filter and target segmentation based on local SNR.The proposed method can achieve more accurate estimates of target size and better background suppression.As a result,it can detect targets with higher detection rate under lower false rate.In Chapter 4,the detection and tracking of complex morphology extended targets are studied.The size and shape of complex morphology small targets can be arbitrary.The joint detection and tracking of complex morphology small targets based on block-wise sparse decomposition is proposed in this work.The detection is mainly achieved by block-wise sparse decomposition within the Principal Component Analysis(RPCA)framework.The image blocks are weighted based on local image complexity and target existence probability.The target existence probability is provided by LMB tracker.The tracking is achieved by improved LMB tracker.Because of exact decomposition,classical 2D target measurements are extended to 3D measurements and additional direction information is provided to improve tracking performance.For tracking,in order to reduce computational cost,the box-particle filter is alsointroduced.Unlike the traditional point-particle approach,the measurements of extended targets are modeled as interval measurements,and the corresponding likelihood function is given based on interval analysis.Then,the LMB tracker and IMM-LMB tracker are implemented by box particles for linear and maneuvering extended targets,respectively.Compared with traditional point-particle approach,the proposed approach can reach a similar accuracy with less runtime.In Chapter 5,the passive ranging method for boost-phase targets and recognition method for midcourse targets using optical sensors are studied,respectively.The estimated ballistic parameters are important output specification.They are very useful for target recognition or intercept.However,they are restricted by the accuracy of ranging.In order to improve the estimation accuracy under single satellite,the passive ranging based on atmospheric absorption of oxygen ?A? band is adopted in this work.The performance of passive ranging under different atmosphere models,weather scenes and angles of view are discussed and simulated by MODTRAN,respectively.For midcourse target recognition,The network is builded to discriminate warheads from decoys based on weighted sequential extreme learning machine.The proposed weighted sequential extreme learning machine can overcome the imbalance problem and meet the need of sequential update.Simulation results show that the proposed method can improve the discrimination accuracy of targets of interest,avoid superfluous computations and reduce computation cost.