CRIMAC PhD student presents a new annotation-free method in acoustic target classification

Ahmet's research introduces a deep learning model designed to automatically extract acoustic features, eliminating the need for manual annotation. Inspired by the Self DIstillation with NO Labels (DINO) model, the study employs a self-supervised learning approach and tests three different data sampling methods. The findings show that the features extracted by this model noticeably enhance the accuracy of machine learning methods in classifying acoustic targets such as sandeel and other fish species.  This research work also shows the potential of self-supervised techniques to improve fisheries acoustics data analysis, contributing to the sustainable management of marine resources.

For further information - visit the Visual Intelligense website