Target Detection And Discrimination Using One Dimensional Range Profile Data

Due to the higher range resolution, most targets are referred as range-spread targets or distributed targets, whose energy is distributed in more than one range unit of the radar echo. For a range-spread target, more information is contained in the echo. It has become an urgent issue in radar technology to use the information effectively. Now, the recognition based on HRRPs has got extensive attention. The detection of range-spread targets is still remained in theoretical research, and the discrimination of detection results is rarely studied, which is used to remove the false alarms in detection results. In this dissertation, the detection of range-spread targets is researched, and a discrimination method based on one dimensional range profiles is proposed. One-class classifiers are combined with the detection results in this method. The main work of this dissertation is as follows:1. The linear frequency stepped chirp signal and statistical models of the radar clutter are studied. Firstly, the criterion of signal parameters, velocity compensation methods, HRRP synthesis methods and their simulation results are given and compared with each other. Secondly, the power spectral models and amplitude statistical models of radar clutter are introduced, and two commonly used clutter simulation methods which are ZNML and SIRP are also given.2. The detection of range-spread targets is researched. First of all, the detection theory of statistical signal, Ney-Pearson criterion and the meaning of CFAR detection are introduced. Next, some CFAR detection methods of point targets and the methods to calculate their detection thresholds are given. The approximately CFAR property of these thresholds is proved according to the simulation results. Then, the detection based on binary accumulation of range-spread targets is introduced. In Rayleigh clutter and Lognormal clutter, some detection methods of range-spread targets are proposed. Their calculation methods of detection statistics and thresholds are also given. Finally, some simulation experiments are done to compare different detection methods.3. Target discrimination based on one dimensional range profiles is studied. In the beginning, the main idea and classification process of the nearest neighbor one-class classifier, the K-nearest neighbor one-class classifier and the K-center one-class classifier are explained with examples. By introducing the Hausdorff distance which can measure the difference between two point sets and improving the nearest neighbor one-class classifier, a discrimination method combined with the detection results based on one dimensional range profiles is proposed. Then this method is spread to the K-center one-class classifier. Based on the measured data, the effect of different classifier parameters on the discrimination performance is analyzed in detail. The performance is also compared between the Euclidean distance and Hausdorff distance. This can prove the effectiveness of the proposed discrimination method. The last, the respective advantages of the improved nearest neighbor one-class classifier and the K-center one-class classifier are compared.