Broadband fish identification of Laurentian Great Lakes fishes

Broadband acoustic echoes were collected on free-swimming alewives Alosa psuedoharungus, rainbow smelt Osmerus mordax, and bloaters Coregonus hoyi. Concurrent midwater trawls were used to determine species composition. A genetic neural network was trained on broadband echoes from each species to an overall correct classification rate of 91%. Tests of the trained network suggested an overall expected correct classification rate of 80-85% and indicated rainbow smelt and alewife echoes were less likely confused as compared to bloater. Application of the trained network to recorded echo data resulted in predicted species compositions that did not correspond well to those observed in the trawls. The classifiers may have been confounded by inclusion of echoes from various species, especially for bloater. A restriction to echoes collected concurrent to trawls or other controlled situations (e.g., pens or smaller lakes with single species) may be needed for building ´clean´ classifiers.