Dynamical classification of hydroacoustic data from a Baltic Sea experiment

The authors analyze the acoustic signatures of a small boat using nonlinear dynamical signal models. Specifically, they discuss the estimation of the parameters of a nonlinear delay differential equation from this data. Using the model coefficients as detection features they implement a Mahalanobis distance-based decision criteria to perform rigorous hypothesis testing. By analyzing acoustic data recorded in shallow water in the Baltic Sea, they compare the performance of the dynamical detector with a frequency band-matched energy detector and show that the former provides increased detection performance. In addition, they demonstrate that the delay differential equation model can distinguish details of the boat's signature, such as the engine RPM, hence it could be used for classification purposes as well.