Multiway analysis in data SONAR classification

This paper presents a data SONAR classification approach that is based on multiway analysis. The passive SONAR system receives the acoustic signals emitted by ships and tries to categorize them as a function of the similarities between ships of the same class. The identification of ship class through the analysis of its emitted signal is a non-trivial task because the signals received from SONAR sensors frequently contain values that represent combinations of different properties of the real world. Systems for modeling the real acoustic signal produced by a ship must be able to remove irrelevant components to obtain the signal's true value.This work uses multiway analysis for both dimensionality reduction and signal denoising to generate a model of acoustic signature that is compact and robust to background sounds. The parallel decomposition method CANDECOMP/PARAFAC is used to eliminate irrelevant information in the class-ship mapping process. The classification model was calibrated and cross-validated on a real dataset. The results showed the effectiveness of the proposed methodology.