The Research On Application Of Multi-Sensor Date Fusion In Target Recognition

With the development of science technology and the needs of modern war, as an intersect subject, multi-sensor data fusion having been draw broad attention and developed at full speed in recent years. Multi-sensor data fusion technology is able to combine every aspect of information provided by several sensors; and can gain more accurate and all-sided information of observation object; and then get accurate and rapid decision-making and judgment.Target recognition is an important component of data fusion technology and also be a key problem in military affairs research field. General definition of target recognition is that makes out the type or attribute of target. Aircraft plays an important role in local war with high technology. In order to assure the dynamic of enemy, make decision analysis; take the prompt reaction and win the war, effectively and quickly recognition of Aircraft target is beneficial to commander. Therefore, the research on aircraft target recognition is very important in academic significance and valuable in practical applying, and it also has the important strategy significance and social effect to the building of our national defense. Many researchers have done a lot of research and exploration in the field both domestic and abroad, but because of the continuous movement of aircraft targets and the constantly changing stance, it become difficult to recognize aircraft in three-dimensional space. Three-dimensional target recognition can be transformed into two-dimensional problems. Therefore, the paper firstly pays attention to the research on aircraft target recognition of two-dimensional, after verifying its feasibility, it extend the algorithm to three-dimensional condition.Conventional multi-sensor data fusion is carried out on the data level, feature level and the decision-making level. From the perspective of multi-level fusion, the paper proposed a kind of target recognition system model with three-level data fusion. The system model fully combines the advantage of D-S evidence theory in aspect of uncertainty reasoning with non-linear process ability of artificial neural network. The paper solved bottleneck problems caused by unable to gain basic probability assignment function using neural network self-learning function, the system model organically combines neural networks with D-S evidence theory. The simulation proves that the new designed system model has higher recognition rate comparing with the one level and two level target recognition system models, and it is suitable for other kind of target recognitions.