A Track Association And Fusion Algorithm In Data Fusion Systems

In this thesis, the data fusion method for the correlation and path state estimationfusion problem are studied for multi-sensor intelligence network in the modern air defensecommand automation system.The correlation algorithms have been divided into three categories according to therelated ideas, which are two-two correlation, multi-dimensional distribution and targetclassification. Consequently, a modified kohonen neural network correlation algorithmbased on the thought of target classification is proposed, which would overcome thedefects of those algorithms, either getting errors or missing expected related track whenthere are cross or bifurcation in the target moving patch in two-two correlation algorithm,or getting explosion where the computation might growth exponentially with theincreasing number of the sensors and the targets in the multidimensional partitionalgorithm. The presented algorithm is made up of four main modules，including clusteringassociation module, target state estimation module, neurons optimization module, and statefusion estimation module. The suitable thresholds for each layer of the competitionelement in the nerves nets are set to avoid the phenomenon of necrotic neurons，which isoften occurred in the conventional method due to inappropriate choice of initial weights.The learning rules are also put forward for the clustering of multi-sensor measurement datain a self-organizing manner. Besides, data clustering association on continuous-time dataare applied to achieve the correlation of moving target tracks. Simulations have been doneto simulate the real flight target track association, which have comparied not only with theother two algorithms, but also with the conventional Kohonen neural network. The resultshave shown the feasibility and efficiency of the proposed modified kohonen neuralnetwork correlation algorithm.The thesis further analysis the weakness in the existing track state estimation fusionalgorithm. On the one hand, previous track fusion problem only considered adaptabilityand completeness aspect of fusion strategy, and regardless the confidence degree of thesensor information. In fact, Even if there are the same recognition rates of the same sensor, it would provide different data due to the interference is different and the credibility of thesensor is different too. Especially in the conditions of more clutter denser and objectivemeasurement data, the credibility of the data source for the validity and reliability of thesystem is more important. On the other hand, the performance of the system and theoptimal ability of the algorithm should admittedly be considered in data fusion systemdesigning, nevertheless the amount of computation, computer capacity and systemcommunication ability and other homologous factors should also be considered. So aweighted data fusion algorithm is proposed to solve the track state estimation fusionproblem, which is built from the engineering practice and practical point of view, to reducethe complexity of the calculation in conventional algorithm. Father more, a kind ofimproved weighted data fusion algorithm based on the accuracy of the set weight in highcredibility sensor premise is puts forward in the thesis, which taking into account thedistribution of the weights depend on the accuracy of each sensor output track. In thismethods, the more accurate initial position center is determined by the subordinaterelations among sensors, which overcome the weakness of those dynamic weightsallocation algorithm in which the centre is not always reliable.Through the simulation of track state estimation fusion, contrast fusion error of thatcredibility have been solved before and after, obtain the high credibility sensors. Weighteddata fusion algorithm, weights dynamic allocation algorithm and improved weightsdynamic allocation algorithm have been used to carry on simulation research that threetarget track fusion specific problems. Simulation results verify the reliability of thealgorithm.