Three-State Self-Organizing Maps Network For Frame Separation Of Sequential Data

The data properties of High Range Resolution Profile (HRRP) are studied at the beginning of the thesis. It is concluded that the HRRP is a kind of sequential data. Then the methods of radar target recognition based on HRRP and their model-database are introduced, in which the frames of the sequential HRRP data are uniformly separated with many drawbacks. What we need do is to separate the sequential data into frames such that we can construct a model-database accordingly that can describe all states of targets and the number of models in it is not large. We separate these radar data into different frames by Self-Organizing Maps(SOM) algorithm, and these frames are used to get a ideal model-database. In separating these radar data into different frames, the conventional Self-Organizing Maps(SOM) algorithm can keep topological ordering of classes, but can't deal with sequential data. In this paper, we present an improved method, three-state Self-Organizing Maps, which is a modified version of the conventional SOM in selecting winner, updating and self-adaptive frame number. The three states are active state, activable state, and died state. Our experimental results on HRRP frame separation show the great effectiveness and efficiency of the proposed algorithm.