Statistical and neural techniques to buried object detection and classification

Systems able to retrieve objects embedded in the sea bottom are of crucial importance for many different tasks, In this paper, an assessment of different advanced signal processing techniques for object detection classification is proposed, Such an assessment based on real data acquired by a parametric sonar, A detector based on the "classify-before-detect" paradigm has been developed that is well suited to exploit the available statistical and spectral a priori information. In other words, a statistical classifier has been employed to distinguish between two classes (i.e., target presence and target absence) exactly like detectors. Concerning classification, the focus has been placed on the exploitation of the object resonances and a neural network has been designed. As a first step the possibility for the network to be properly trained has been assessed, while as a second step the discrimination capability has been verified. The above methods have been tested with an experimental data set in which the echoes of a steel cylinder were embedded in real reverberation echoes, according to low signal-to-reverberation ratios.