Application Of Rough Sets In Target Recognition With Data Fusion

This thesis was based on target recognition techniques in information fusion. In consideration of that the information of the same target from a single sensor had relative big uncertainty and the recognition performance using single sensor was not good, we explored efficient methods and realization techniques of acquiring useful information and extracting knowledge using multi information sources. And we also discussed how to use multi-source information to increase target recognition rate. In the process, rough theory and the technique of fusing rough sets and neural networks were introduced, we focused on information extraction, knowledge representation and simplification, property reduction, rules modeling, increment learning and rough neural networks based on BP and RBF structures in information fusion. To deal with these problems, this thesis did the following work: Researched the knowledge simplification and representation system based on rough theory including the basic concepts, property reduction, rule creation and simplification among which simplifying current properties is a focus. In consideration of the need of real time image processing, we proposed a quick property simplification algorithm to overcome the defect of high time cost. With regard to the problems encountered in multi-sensor fusion and the deficiency of current techniques, we proposed to use rough sets to build the model of rules in target recognition works contained in data fusion. Under the situation of no object model or prior knowledge, we created the rules of fused objects based on the original data and built the corrosponding knowledge database. Also we used confidecne and support to evaluate the classification ability of the rules. Since in the process of target recognition, the demand of real time work was highlighted while the image data were quite large, we proposed to add increment learning during rules modeling which speeded up the algorithm and thus increased efficiency. Based on the their separate properties, the relation of rough theory and neural networks was researched and the learning mechanism in neural networks was added to rough set system; also the net work structure was created through the properties of the condition and decision-making in rough sets. Then the rough neural network based on BP structure was researched and a rough neural network algorithm with RBF structure was proposed to overcome the deficiency and limitation of BP network. By fusing rough sets and neural networks, they complemented each other's deficiency and took good advantage of their separate predominance, thus increased the efficiency of the algorithm and recognition rate while decreased false alarm rate.