Rapid surveying of buried targets by hybrid seafloor image and biomimetic signal classification

Acoustic detection and classification of buried mines presents a challenging and, as of yet, unsolved object recognition problem. Techniques for detecting and classifying mine-like targets in backscatter images created with broadband sonars are starting to yield promising results. These images represent energy vs. time mappings of target echoes; however, further information on potential targets can be extracted from the spectral and temporal content of these broadband signals. An approach to classifying the time-frequency characteristics of target echoes measured with a 3D, 5-23 kHz, forward/aft sweeping, buried object imaging system is described. Matched spatial filters are applied to backscatter images representing horizontal seafloor slices. Detection outputs are used to identify beams and ranges for candidate targets. Time- spectral features are then extracted from the appropriate time-series component, and input to an artificial neural network trained to identify mine-like objects. Fusion of image and signal classification algorithms applied to these data is expected to result in rapid identification of buried targets and a reduction in false alarm rates.