Interference rejection, motion compensation, and classification of multi-aspect broadband sonar echoes

Multi-aspect sonar echoes can be classified using features extracted from synthetic aperture images (image-based features) and features extracted directly from sonar echoes (acoustic features). Both feature sets are improved by interference rejection techniques that are specifically designed to remove snapping shrimp sounds and by identification of aspect-and-target-tolerant features that can be used to estimate range shifts between echoes for object registration (motion compensation). Image-based features should include features that depend on the evolution of incomplete SAS images during image construction and features from generalized SAS images that represent attributes other than reflectivity (feature images). A proposed new imaging algorithm uses a combination of aspect changes (as in SAS) and beam scanning (as in beam pattern deconvolution). Acoustic features are divided into aspect tolerant features that can be used with a single echo, transition matrices that describe range-and-aspect dependent transitions between vector quantized segments of a feature-gram (a generalized spectrogram), and multi-aspect features that are not defined for a single echo. The proposed tasks are to develop algorithms to implement and test all of these interference rejection, motion compensation, and classification concepts, and to combine them into a powerful classifier for multi-aspect broadband sonar echoes.