Passive tracking and detection of underwater narrow-band acoustical spectral signatures

Tracking, detection and classification of targets is complicated in presence of multiple targets located at close or overlapping bearings. Assuming that the respective targets exhibit unique spectral signatures, which include narrow-band (NB) tonals and associated harmonics, this problem is addressed in several steps using conventionally beam-formed array data. The first two include detection of signals and associated bearings, followed by clustering bearings and associated frequencies. Signal and bearing detections are carried out independently analyzing their time-frequency distributions sorted along bearing in direction of descending spectral power. Values of power and bearings are grouped independently from each other into vectors, which in turn form sets called time-frequency (TF) cells. The Maximum Mean Discrepancy (MMD) test of empirical centres of masses of respective TF cells in an inner product space is a similarity measure used for required detections. Respective MMD's of power and bearing TF cells are used to obtain weights used to cluster detected bearings and associated frequencies. A signal detected at a given frequency is used to start a single frequency-bearing Kalman tracker (SFBT). A multivariate frequency-bearing tracker (MFBT) is started when the SFBTs associated by a common bearing are observed repeatedly at relative frequency exceeding a predefined threshold. A state vector of MFBT is propagated in time if detections are observed at least at two out of N associated SFBTs. For each SFBT frequency gates of few per cent of central frequency are used. Inconsistency check of latest MFTB bearings observed at different frequencies is used for MFTB stopping decision. Along bearing of each MFTB a mean spectrum is calculated. This spectrum corresponds to a target signature, which can be used for classification purposes in future.