Automatic fish school classification for acoustic sensing of marine ecosystem

With the human demand for fish and the global warming effects, we know that marine populations are changing. Developing methods for observing and analyzing the spatio-temporal variations of marine ecosystems is then of primary importance. In this context, underwater acoustics remote sensing has a great potential. Operational systems mainly rely on expert interpretation of echograms acquired by sonar echosounders. In this works, we propose new algorithms for the analysis of acoustic survey regarding the inference of species mixing proportion. They rely on the definition and training of probabilistic school classification models from survey data.