Annotation-Free Feature Learning for Improved Acoustic Target Classification

Abstract: Acoustic surveys are an important source of data for fisheries management. During the surveys, ship-mounted echo sounders send acoustic signals into the water and measure the strength of the reflection, so-called backscatter. The collected backscatter data are then annotated and utilized for Acoustic Target Classification (ATC) tasks, which involve identifying backscatters and classifying them into specific categories such as sandeel, mackerel, and background (including bottom and plankton). However, the process of annotating data is typically manual, making it a labor-intensive and time-consuming task. The main goal of this study is to create a deep learning model that can extract acoustic features without the need for annotation, thereby enhancing the representation of acoustic data. To achieve this, we employ a self-supervised method inspired by the Self DIstillation with NO Labels (DINO) model. This model is trained using three distinct data sampling methods, and the quality of the extracted features is subsequently evaluated and compared. Our results demonstrate that the features extracted by our model enhance the discriminative power of various machine learning methods for ATC tasks when compared to the untreated data. Our findings highlight the advantage of applying emerging self-supervised techniques in fisheries acoustics. This study thus contributes to the ongoing efforts to streamline and improve the efficiency of acoustic surveys in fisheries management.

‍

In compliance with GDPR consent requirements, presentations given in a Visual Intelligence context may be recorded with the consent of the speaker. All recordings are edited to remove all faces, names and voices of other participants. Questions and comments by the audience will hence be removed and will not appear in the recording.  With the freely given consent from the speaker, recorded presentation may be posted on the Visual Intelligence YouTube channel.

‍