Target Detection In Sea Clutter Based On ELM

In the study of all aspects of the ocean, the research of sea clutter has been given more and more attention by scholars who have carried out in-depth analysis of its statistical characteristics, time-frequency distribution characteristics, fractal characteristics and any other characteristics. Based on these characteristics, scholars put forward a number of target detection methods in sea clutter.This thesis begins with the introduction of IPIX radar measured data in 1993, which lays the foundation of the later experiments. Then through the sea clutter’s statistical characteristics, time-frequency characteristics and fractal characteristics, we study the time and space correlation, the amplitude distribution fitting, one and two dimensional simulation, FRFT (Fractional Fourier Transform) domain and fractal feature modeling. As a result, the de-correlation time, the K distribution fitting parameters and the Hurst index of fractal modeling on FRFT domain are proved to be three feature vectors that can better distinguish target echo from sea clutter. Finally, the thesis introduces the theory and algorithms of Extreme Learning Machine (ELM), and its advantages compared with traditional classification methods. Extreme Learning Machine is used as the classifier to realize pattern classification. Based on the radar sea clutter data experiments, it shows that the method of target detection in sea clutter with ELM can distinguish target echo from pure sea clutter effectively, and reduce the computation time significantly, which can meet the requirements of real-time processing of sea clutter data.This thesis uses de-correlation time, K distribution fitting parameters, Hurst index of fractal modeling on FRFT domain as feature vectors and takes advantage of Extreme Learning Machine for classification, which has a certain theoretical and practical value on the study of target detection in sea clutter.