The Study And Application Of Multi-sensor Data Fusion

With the rapid development of modern technology and computers, multi-sensor data fusion technology has been widely used in modern military and civilian, also has shown tremendous research potential. The application which is more and more used in the military has become the focus of national defense. Modern war is a science and technology war. Data fusion plays a huge role in determining the target track and identification.The information given by sensor includes location information and identity information. The known location information can predict the location of the target under test by track association method. Track correlation methods include nearest neighbor data association method, the probabilistic data association method and the joint probabilistic data association method. Joint probabilistic data association method is an improved method based on probabilistic data association method. The known identity information can predict specific identity of the target under test by technologies of target identification and integration. Target recognition technique includes the similarity coefficient method, statistical pattern recognition technique, neural network method and so on. Neural network which rises in recent years is a new method. It has more complete theory and has great research potential, which is widely used in many applications. Diagonal recurrent neural network (DRNN) is a kind of feedback neural network, can be used in prediction and classification. Identification fusion technology includes the best method of target identification integration, DS evidence theory and so on.Researches on data integration at home and abroad are also increasing, particularly in the identity integration. In the D-S evidence theory, as Dempster combination rule has its own shortcomings which can not correctly fuse the conflict evidences, many improved methods have been proposed. On the basis of these methods, the paper introduces pignistic probability function, proposes an improved method based on pignistic probability function.Through specific examples of this method, we can see this method has small computation, fast convergence and can give more rapid and accurate synthesis results. We apply this method to the gear fault detection system. By specific examples, performances of each method have been compared and analyzed, experiments show that the method proposed in the paper can accurately give the identity information of target under test.