Acoustic identification of marine species using a feature library

TLDR

Sonars and echosounders are widely used for remote sensing of marine life, yet acoustic identification of species remains uncertain, necessitating more accurate and objective methods. The study aims to enhance acoustic species identification by processing multi‑frequency echosounder data through a modular pipeline that detects schools and classifies targets using the Large Scale Survey System software. This pipeline employs overlapping acoustic beams, a feature library of relative frequency responses, and translates categorization results into species abundance estimates, and is applied to data from the Barents, Norwegian, and North Seas. Automatic categorization closely matched manual labeling across all surveys, particularly for schools, indicating high conformity.

Abstract

Sonars and echosounders are widely used for remote sensing of life in the marine environment. There is an ongoing need to make the acoustic identification of marine species more correct and objective and thereby reduce the uncertainty of acoustic abundance estimates. In our work, data from multi-frequency echosounders working simultaneously with nearly identical and overlapping acoustic beams are processed stepwise in a modular sequence to improve data, detect schools and categorize acoustic targets by means of the Large Scale Survey System software (LSSS). Categorization is based on the use of an acoustic feature library whose main components are the relative frequency responses. The results of the categorization are translated into acoustic abundance of species. The method is tested on acoustic data from the Barents Sea, the Norwegian Sea and the North Sea, where the target species were capelin (Mallotus villosus L.), Atlantic mackerel (Scomber scombrus L.) and sandeel (Ammodytes marinus L.), respectively. Manual categorization showed a high conformity with automatic categorization for all surveys, especially for schools.