Robust signal processing algorithms for detection of small targets in sea environment

Abstract

A robust scheme is introduced and analyzed for its performance to detect small targets in sea environment. This scheme includes the following special features:(i) The phase space of the system is extracted from the observed time series using embedding delay method, followed by Kernel Independent Component Analysis (KICA) algorithm as a feature selector. (ii) The prediction model is constructed using real time recurrent learning (RTRL) neural network.(iii) The stability is maintained in case of recursive prediction using the largest Lyapunov exponent of the system. The algorithms are studied in detail using Lorenz map and five different sea cutter time series datasets.