Detection of objects buried in the seafloor by a pattern-recognition approach

Systems able to retrieve objects embedded in the seafloor are of crucial importance for many different tasks. An experimental assessment of a detector applying the "classify-before-detect" paradigm is proposed. The evaluation is based on real data acquired, during two sea trials, by two different sonar systems using low grazing angles and placed far from a target object. The "classify-before-detect" paradigm is a pattern-recognition approach to designing a classifier aimed at distinguishing between two classes (i.e., target presence and target absence), just like a detector. This approach has been selected and developed as it is very well suited to exploiting the available statistic and spectral a priori information on the target echo. In short, some features are extracted from the Wigner-Ville distribution and the bispectrum of partially overlapped short segments of the acquired echo signals. The dimensionality of the problem is reduced by the principal-component analysis, and the reduced feature vector is sent to a supervised statistical classifier. The ideal training set is composed of pure reverberation signals and the responses of the target in free field at different aspect angles.