Résumé de la publication

The commercial availability of a calibrated broadband echosounder system operating at fisheries acoustic frequencies on a variety of platforms (vessels, moorings, gliders, drones …) has led to an increase of broadband fisheries acoustic data collection. Broadband acoustic images (BAI) provide detailed spectral information which can be used to classify a larger variety of scaterrers. However, their handling and analysis poses computational, analytical (‘curse of dimensionality’) and validation challenges. Building upon the hyperspectral image community legacy, we present the first results of unsupervised classifications of BAI collected near an offshore windmill in the Bay of Biscay (BoB, France). BAI were produced by echo-integrating broadband fisheries acoustic data on a fine mesh grid, using the pymovies_3D Python package. Shallow dimension reduction and classification techniques were tested on a reference dataset comprising spectral broadband signatures of scatterers identified in the BoB. Classification techniques performing well on this standard reference ‘echoscape’ were then applied to classify in-situ BAI, in an attempt to detect areas where scatterers form groups with similar spectral properties. Mean frequency spectra of identified clusters will be compared to scaterrers models and biological sampling data collected in the area, in an attempt to characterise the local pelagic ecosystem.