Study Of Several Data Fusion Algorithms Based On Targets Recognition

With the comprehensive applying of information technology, the technology of multi-sensor data fusion is obtaining great development, and studies of methods to multi-sensor data fusion have been the key point. This paper introduces the elementary theory of data fusion and several useful algorithms, which are often applied to recognize targets in data fusion areas, and as followed the fundamental of these issues; we have studied and analyzed especially on Bayesian, D-S evidence theory and neural networks, in additional, D-S evidence theory and neural networks method have been combined in order to recognize targets.D-S evidence theory is a complete method, which is able to solve out the uncertainly issues effectively. At the same time, there are also faults of traditional D-S evidence theory, such as conflict of evidence proofs, great amount of computation, irrespective evidence and so on. However, it is a quite strong request of irrespective evidence, which can not be satisfied strictly in many situations. Considering those above, this paper put forward an amendatory method of D-S evidence theory, which has both important theoretical value of studying and good applied value.According to it's recently development, neural networks has been applied comprehensive in many fields, and a sort of multilayer forward networks with unilateralism spread has become the one of the BP neural networks, which has been applying more universally. Although BP model has its important significance in every aspect, the problems of local minimum with a slow speed of convergence are still existed in the model. In this paper, a combination algorithm between BP neural networks and fuzzy reasoning was introduced to recognize targets effectively.Considering these questions of low accurate identification, bad stabilization and solution of uncertainty in some ways of multi-sensor system at present, a new algorithm of date fusion combined D-S evidential theory with neural networks is put forward which we call it DSBP algorithm. According to the characteristic of characteristic information that the multi-sensor obtained, DSBP algorithm divides it into some groups and set up a corresponding neural network to every group, at the same time we introduce a concept of unknown probability to the goals based on the result of credible probability of these goals, at last we have a fusion of time and space depending on the transposition result of the neural networks'output by D-S evidential theory. This method has the advantage of solving out the uncertainly issues effectively of D-S evidence theory algorithm and that of pattern reorganization of neural networks, and solves the problem that the general ways of data fusion can not identify the multi-sensor's uncertainty information of great noise at present. The simulation shows that the way can effectively improve the rate of the targets'identification and great antinoise capacity.