Can acoustic shadows identify fish species? A novel application of imaging sonar data

Abstract

This study addresses a fish species discrimination method based on normalised elliptic Fourier descriptors applied to acoustic shadows derived by Dual-frequency Identification Sonar (DIDSON). Acoustic shadows of templates (20, 30, 40 and 50cm) and live fish of four species [bream, Abramis brama (L.); barbel, Barbus barbus (L.); chub, Leuciscus cephalus (L.); and trout, Salmo trutta (L.)] were projected on a plate in an experimental set-up and tested on suitability for species discrimination. Twenty-centimetre templates were correctly classified in 97.5% of the cases, indicating a size threshold. The larger templates reached values of 100% correct classification based on cross-validated discriminant function analysis. It was also possible to classify moving fish based on screenshots of their acoustic shadows with a certainty of 83.9%. Extended field tests are required to evaluate the method for use in practical monitoring applications in multispecies river environments. © 2012 Blackwell Publishing Ltd.