Passive signature characterization and classification by means of nonlinear dynamics

Abstract

We analyze sonar recordings of various boats as well as ambient sea noise using nonlinear dynamical signal models. Specifically, we discuss the estimation of the parameters of nonlinear delay differential equations from data. Using the model parameters as classification features we implement a three class Bayesian minimum-error-rate classifier and demonstrate almost perfect classification of the data set considered. This indicates that classifiers based on nonlinear dynamical models can be useful in sonar applications.