Methods Of Multi-Sensor Data Fusion And Their Application In The Spatial Targets Recognition

Multi-sensor information fusion theory and its applied technology have obtained rapid advances of late years, which becomes an important research field, and now, recognition method based on information fusion is a hot spot of research. For being the key technology of spatial surveillance, spatial target recognition is a very important research direction. For the characteristic of spatial target recognition, multi-sensor data fusion is likely to break the choke point of traditional method. This paper studies the methods of multi-sensor data fusion, and applies these methods in recognition of spatial targets.The primary task is that how to judge the information from defferent sensors belongs to one target or one kind of targets in spatial target recognition, we regard this problem as track-to-track correlation. In present methods of track-to-tack correlation, the false and lost track-to-track correlation have not been taken into account in the complex background with dense targets. So two kinds of methods are proposed to deal with this problem, one of which is the correlation algorithm based on fuzzy synthetic decision and D-S evidence theory, another is based on K-nearest neighbor (K-NN) principle and D-S evidence theory. The two methods combine the logic of fuzzy decision and strictness of statistical classification with intelligence of evidential theory successfully. Simulation results show that both methods can solve the problem of fault and lost correlation well.Base on correlated information of targets, the next task is how to recognize the identities of targets using multi-sensor information fusion. We consider the method of neural network for it having been used in this field effectively. However, single neural network has some faults such as bad stability, bad convergence and so on. It solves these problems by using neural network based on fuzzy decison and neural network group. Compared with traditional network, neural network based on fuzzy decison has simple structure, clear logic layer and short training time, while for network group, it is more intelligence and fuses uncertain information better without longer training time. Simulation results show that neural network based on fuzzy decision can get a satisfying recognition result, and group of neural networks not only diminishes the training time of network, but also improves the efficiency and...