Research On Techniques Of Adaptive Data Fusion

The overall performance of data fusion systems is affected by the nature of the sensors, the operational environment and the targets, the adopted fusion models, process, and so on. The dynamic changes of the nature of the sensors, the operational environment and the targets lead to that the data fusion systems with changeless fusion models, fusion process and fusion parameter wouldn't get perfect results. Enhancing the adaptability of data fusion systems and weakening the effect of outside factors on the performances of fusion system is a hot research problem. This thesis thus focuses on the improvement in the adaptability of data fusion system. The idiographic research is as follows:The evaluation of the track correlation uncertainty is researched. The concept of the uncertainty of track correlation is defined based on the uncertainty theory. The evaluation method for track correlation uncertainty is revised based on the measurement uncertainty. The revised method can be comprehended and implemented more easily.An adaptive method for data correlation is presented and implemented. The length of sequence disposal usually is set changelessly. With the evaluating of track correlation uncertainty, the length of sequence is made to be self- adjustable, which makes the decicision of the data correlation more timely and more reliable, and so improved the adaptability of data correlation considering the different sensor accuracy, target distribution and routes.A sensor bias estimation method is presented and implemented with the base of data driven control theory. The traditional estimation methods often have large residual errors. Based on iterative learning approach, a data model of bias estimation is implemented with the use of the high precision positioning information of the collaborative targets.The architecture and process model of an adaptive data fusion system is designed, the corresponding software is developed and the simulation experiments is made. The experiment results show that the data correlation can adjust the length of sequence disposal according to the sensor accuracy, the target distribution and routes, the bias estimation method can acquire less residual bias and the data fusion system thus achives the capability in adapting the change of the nature of sensors and targets.