A Multispectral Target Detection in Sonar Imagery

Detection of underwater objects in sonar imagery is a key enabling technique, with applications ranging from mine hunting and seabed characterization to marine archaeology. Due to the non-homogeneity of the sonar imagery, the majority of detection approaches are geared towards detection of features in the spatial domain to identify anomalies in the seabed's background. Yet, when the seabed is complex and includes rocks and sand ripples, spatial features are hard to discriminate, leading to high false alarm rates. With the aim of detecting man-made objects in complex environments, we utilize, as a detection metric, the expected spectral diversity of reflections to differentiate man-made objects' reflections from the relatively flat frequency response of natural objects' reflections, such as rocks. Our solution merges a set of preregistered sonar images, each of which are obtained at a different frequency band. Using the Jain's fairness as a metric to evaluate the spectral diversity of a suspected object within a low or high resolution sonar imagery, respectively, our solution detects anomalies across the spectrum domain. We tested our algorithm over simulated data and over multispectral data obtained in a designated sea experiment. The results show that, compared to benchmark schemes, our approach obtains better performance in terms of the trade-off between false alarm rate and detection capability.