Multi-sensor Air Target Identification Method

In the future war field, in order to obtain exact and comprehensive information in time, using single sensor can't meet the demand, so we must replace them by multisensor (radar, infrared searching and tracking system, estimation of friend or foe identification, inner data link for fighter plane and so on). Multisensor target identification technique has become one of key techniques in modern fire control system.This thesis mainly introduced the theoretical basis of data fusion and Relevance theory. Based on this, key introduced neural network and D-S evidence theory, analysed their characteristics, then designed a new algorithms which integrate the advantages of D-S theory and neural network. Due to neural network has distributed memory and parallel processing way, self organization and self study function, and very strong merit such as fault-tolerant and robust capability. Thus it has got more and more broad application in the field of pattern recognition. The paper priority introduced BP neural network, this network has strong nonlinearity mapping ability, adopt a method that the overall situation approaches, have a fairly good fault-tolerant ability. Therefore, this paper designed a new algorithms based on evidence theory and BP neural network, which integrate the advantages of D-S theory and neural network. Thereby, improve the systematic stability and distinguish probability, more effective carried out the target identification.The main work of this paper is: After carrying out a lot of research on the basis of theory, A algorithms given based on Dempster-Shafer evidence theory and neural networks, in view of the defects of BP, given Several improving algorithm. This paper designed a Multi-Sensor Target Recognition System, designed a new algorithms which integrate the advantages of D-S theory and neural network. The result is use multi-sensor data fusion system is much better than a single sensor, the stabilities and identification system are both improved a lot after combing two kinds of methods.