Ship Targets Association Based On Satellite Electronic And Imaging Reconnaissance Information

Target association is one of the most pivotal pretreatments in the multi-sourceinformation fusion, which aims at determining whether the multiple reconnaissanceinformation is from the same object or not, providing the sub-objects and fusionsubstance for high level information fusion and ulteriorly reducing the confliction andredundance information, thereby obtain the more integrated description of the targets.As the kinds of the most important mobile military targets, the correspondence of theship targets within multiple observational data can be determined by target associationand hence provide the supportment for the following object validation and tracking.Within the background of the maritime surveillance and satellite reconnaissance, thedissertation studies the ship targets association algorithms based on electronic andimaging reconnaissance data. The main contributions of the thesis are summarized inthe following.(1) The paper presents the scientific signification and application value of targetassociation which based on electronic and imaging reconnaissance; reviews the relatedresearch work and the future development of target association. The crucial researchcontents of the dissertation are pointed out finally.(2) In the research of the ship location information extraction, the dissertationfirstly introduces three classical extraction methods for imaging reconnaissance data andmainly analyse the extraction method for electronic reconnaissance information. Due tothe fact that electronic reconnaissance can be utilized to observe the given target inscout region continuously, thus the problem of location information extraction byelectronic reconnaissance information can be changed into the problem of track initial.Currently, track initial algorithms are generally applicable to the periodic andcontinuous active reconnaissance which has a high time resolution. Whereas theelectronic reconnaissance data are commonly aperiodic and uncontinuous, a new trackinitial algorithm based on cascaded clustering method was proposed to acquire thetracks of ship targets. Experiments on both simulated and real world data showed theproposed method can be used to resolve the complicated track initial problems when thenumber of ships, dynamic status and other prior information are unknown.(3) In the research of target association methods based on topologicalcharacteristics, due to the long revisit periods, the dynamic model of moving targetscannot constructed exactly, thus the traditional algorithms which based on the locationinformation cannot be utilized directly. In order to solve the problems above, the pointpattern matching method can be used to achieve the target association. On account ofthe dissimilarity of location precision of the two different sensor data, the thesisproposed a new shape descriptor named Point Pair Topological Characteristics (PPTC) to describe the relationships of relative locations between ships. Contrapose to theorientation error and outliers of electronic reconnaissance data, after combining theinvariant feature with probabilistic relaxation labelling and spectral matching methods,the dissertation presents the one target association algorithm which based on PPTC andprobabilistic relaxation labelling (PPTC-PRL) and the other target association algorithmwhich based on PPTC and spectral matching method (PPTC-SM) respectively.Experiments on both simulated and real world data showed that the two proposed novelmethods all are more robust to noise and outliers compared with the classical methodsunder similarity transformation. The PPTC-PRL method is more robust to outliers thanthe PPTC-SM method and inversely the PPTC-SM method is more robust to noise thanthe PPTC-PRL method.(4) In the research of target association based on topological and attributivecharacteristics, for the problem of the attributive characteristics cannot apply toassociation directly by reason of the two reconnaissance data describe targets indifferent ways, the thesis firstly proposed an target recognition algorithm based onRough Set and Radial Basis Function Neural Network (RS-RBFNN). The RS-RBFNNmethods can unify the different attributive description and therefore recognize bothoptical imaging targets and radars targets. For improving the accurate association rates,a new target association algorithm based on D-S evidence which synthesize thetopological and attributive characteristics together was proposed by using thecomplementary of the topological and attributive characteristics,. Experiments indifferent scenarios showed that, compared with the other target association algorithmswhich only relied on topological or attributive characteristics, the proposed method canimprove the accurate association rates to some extent and is more robust to locationerror and outliers.