Target Classification Of HF Over-the-horizon Radar Based On Multiple Frequency Feature And Poles

The main motive of this thesis is to investigate target classification of HF over-the-horizon (OTH) radar based on multiple frequency feature and poles. The HF OTH radar can detect aircraft and ships in the area of more than thousands of nautical miles beyond the horizon. It is an efficient way to discriminate targets utilizing multiple frequency feature and poles. The main jobs and contributions in this thesis are as follows:Linear discriminant method dose not depend on condition probability density function estimated from known data. The application of linear discriminant to classify simple and complex plane objects is investigated utilizing multiple-frequency data as feature vector. This approach can only solve the problems that are linearly separable.The misclassification probability of the nearest neighbor decision rule won't exceed 2 times of that of Bayes decision rule when the sample number is very large. The classification results of simple and complex objects are presented utilizing the nearest neighbor discriminant. The results proved that the nearest neighbor rule can solve the problems that can't be separable linearly.Artificial neural network has the capability of self-organizing and self-learning .It can accommodate to the environment and deal with many kinds of complicated and random information. The classification of simple and complex objects is investigated using the multiple layer forward neural network and the self-organizing feature map network .The problem of estimating the pole number of radar targets from transient impulse response in noise is not resolved perfectly by now. A very efficient information criterion function means——MDL is presented to estimate the number of poles of the targets.Prony method is most sensitive to noise when it is used to extract poles of objects. The state space method is more meaningful and understandable, it can be managed flexibly.The two approaches are both used to extract the target poles. The results showed that the state space method is more efficient and more insensitive to noise.