Spectral discrimination of fish shoals from seafloor in the Gulf of Maine during the ocean acoustic waveguide remote sensing (OAWRS) 2006 experiment

Abstract

In this thesis, we perform a spectral discrimination of fish shoals from background returns using statistical techniques. Classification of fish species requires an efficient and solid approach to distinguish fish scattering from seafloor returns. Neyman-Pearson Hypothesis Testing, Kullback-Leibler divergence, Matched Filter and discriminating based on the shape of the spectral dependence, methods originated from Detection theory, are applied in well documented cases from Gulf of Maine during spawning season to distinguish seafloor returns from fish scattering across frequency domain. The discrimination of fish shoals from seafloor returns is achieved by analyzing the absolute levels of scattered returns and the pattern of their frequency response. A generalization of the statistical techniques is developed that enables all frequencies to be tested at once, allowing the spectral discrimination and echolocation of fish shoals from regions dominated by background returns. Conclusions derived from statistical techniques are consistent with physical evidences, such as in situ echosounder measurements and frequency responses. Fish shoals are distinguished from background regions by evaluating the likelihood ratio test, matched filter and analyzing the slope of the frequency dependence of all pixels in an examined ocean acoustic waveguide remote sensing (OAWRS) image.