Buried underwater target classification using frequency subband coherence analysis

This study introduces a new feature extraction method for detection and classification of buried underwater mine-like objects. Multiple sonar returns off an object are used, where each sonar return is characterized by its specific frequency subbands, which contain valuable discriminatory information that can be used to correctly determine the type of object encountered. Features are extracted from the data using canonical correlation analysis (CCA) between the selected frequency subbands of two sonar returns and are subsequently used to classify mine-like and non-mine-like objects using a simple classifier. This method offers a more rigorous way of performing acoustic color processing where ping-to-ping coherence between sonar returns is also exploited in extracting acoustic color features. This method will be tested against a previously developed time domain multiple aspect feature extraction method on a database that contains sonar returns from various buried or proud mine-like and non-mine-like objects in different operating and environmental conditions. Results will be presented in terms of the receiver operating characteristic (ROC) curve for each system and overall detection and classification performance in different operating and environmental conditions.