Research On Feature Selection And Pattern Classification Method Based On Radar Returns

Nowadays, the trend to extend the information processing and target recognition abilities of radar systems has become more and more important and begins to dominate the development of radar systems. However, because of the complexity of environment factors and the variety of targets and their behaviors, there are lots of difficulties to design a radar target recognition system.Researches in this paper are based on a ship target recognition systems for sea surveillance radar. A greedy feature selection algorithm has been employed, and a wieldy sample selection algorithm basing on the frequency histogram re-sampling has been proposed, which reduces the influence of redundant features and samples, and consequently improved the robustness of the resulting*recognition system. A structure-adaptive network basing on Singular Value Decomposition (SVD) and discrimination-criterion has also been advanced, which could deal with the complexity of recognition patterns and large training-sets, and formulate structure adaptive networks in actual applications.Experiments and performance analysis on the structure adaptive network based on the training-set, which generated by the sample selection algorithm, and the features fetched by the feature selection algorithm, show that the methods-lead to an excellent results under complex patterns and large sample sets recognition application. Although the methods proposes in this paper are based on the analysis of actual radar data, the ideas of these algorithms are not build upon concrete applications and valuable for generalization.