Research On Precession-identification Based On Sparse Representation

As a fine feature of targets, micro-motion is of important value in the field of targetdetection and recognition, and has drawn great attention from researchers abroad andhome. Micro-motion identification is the key and difficult problem in the targetrecognition using micro-motion characteristics, involving the judgment of micro-motionexistence, the determination of micro-motion modes and the estimation of themicro-motion parameters. Under the background of the target recognition in midcourseballistic missile defense, and around the scientific issue of micro-motion identification,this dissertation researches on the sparse representation theory aimed at precessionidentification, the method of precession identification on the sparse representationdomain, and the performance analysis of precession identification. The dissertation is asfollows:1Firstly the applications of micro-motion feature in the midcourse ballistic missiletarget recognition are reviewed, and the classic micro-motion modeling methods aswell as the technologies of micro-motion feature extraction and parameter estimationare summed up. After that, the existing methods of micro-motion mode identificationare introduced. According to the basic characteristics of micro-motion identificationissue, the dissertation states the potential advantages and problems insparse-representation-based micro-motion identification methods.2The ballistic midcourse target motion model is given to discuss thecharacteristics of the micro-motion, based on which, the radar echoes of precessingtargets is deduced. Analysis is made about the micro-Doppler of the spinning symmetrycone target with precession, and comparison is made between the theoreticallycalculated micro-Doppler and the Short Time Fourier micro-Doppler of the simulatedradar echo.3A micro-motion identification framework based on the sparse representation isbrought in, for which the dictionary designation, thel psparse model based onregularization method and the numerical optimization for solution are introduced indetail. The framework applied to the precession identification of midcourse targets, theprecession identification based on sparse representation is studied. On the detection ofprecession modulated signals, a detection process is proposed. The process is based on adata-adaptive threshold, whose factor is determined by a certain tolerance of false alarmprobability and the missing detection probability; when the signal-to-noise ratio is0dB,the detection probability reaches0.955. On the parameter estimation of precessionmodulated signals, the estimation is derived from the relation between the parametersand the dictionary; the method is compared with the time-frequency-extended Houghtransform, and shows a much better performance in parameter estimation. 4In order to improve the performance of the identification based onl pmethods,some researches are done on the sparse representation methods in the precessionidentification of midcourse warhead targets. Firstly, to cope with the difficulty in theselection of the regularization parameter, a regularization parameter selection methodbased on information criterion is studied. It is data-adaptive and shows a betterperformance in sparse solution than traditional methods. Secondly, in order to overcomethe drawback in the identification performance under a low signal-to-noise ratio, thereweighted iterativel palgorithm is proposed. The algorithm can not only suppress thenoise but also enhance the weak signal components. Besides, when the regularizationparameter is not optimally set the, the optimal solution could be attained by feweriterations.5Some performance parameters are proposed to evaluate the precessionidentification performance. On the detection of precession modulated signals, thefollowing parameters are defined: the detection probability which is the probability ofthe existence of precessing targets; the inerrant detection probability which is used tomeasure the freedom of unwanted detection and errant detection of signal components;and the complete detection probability which is the probability of accurate detection ofsignal components. Among those parameters, the inerrant detection probability and thecomplete detection probability are more refined and stricter measurements for thedetection performance. They reflect the ability of the sparse-representation-basedmethods in the detection and extraction of signal components. On the parameterestimation of the precession modulated signals, the following parameters are defined:the normalized mean error which reflects the absolute performance of the parameterestimation and the normalized mean square error which reflects the relativeperformance. The former is used to measure the quantization error between theestimation value and the true value, while the latter is used to measure the fluctuation ofthe estimation value around the true value.In the end, the main contributions of the dissertation and problems in micro-motionidentification to be resolved are concluded.。...