| It' s very important now to research non-cooperate target recognition (NCTR) in modern information war, Neural Network (NN) can be widely used in this field because of its strong self-adapted learning ability and high speedily parallel calculating ability. This thesis researches how to use NN technologies on NCTR . The content and innovation show as follows:1. The primary principle of NN is described, and three typical NN arithmetic and their mathematic models of BP, RBF and Kohonen is introduced respectively.2. Some improvements in Kohonen arithmetic is brought up, such as using neighbor function covering for neighbor circle to ensure fair competition, network training method based on total mapping and network constringency method based on Double mapping. The differences between Kohonen and RBF in architecture and processing are analyzed and the view that they can learn form each other' s strong points to offset one' s weakness is emphasized , two mixed neural networks, KR I and KR II , are set up .3. BP and RBF pattern recognition approach for digital modulation signals based on the statistical parameters histogram distribution are presented, the performance of BP or RBF is analyzed.4. A new clustering method based on single Kohonen neural network resolving data association is brought up in this thesis, this method is used to associate those unknown data automatically.5. A new Combination Neural Network (CNN) method to identify those radiant points data of multi-sensors is brought up. This CNN is composed of four kind of networks: RBF , Kohonen, KR I and KR II . Estimated posterior probability of CNN is given using linearity regression method, which is used as decide-making to realize targets recognition.6. A new confidence measure is presented , it as threshold value is used to reject interferential data .The simulation results show that our approaches are able to realize the NCTR in various environment and can get better recognition effect than single network .
|
PDF Full Text Request