Underwater Fish Classification of Trout and Grayling

Abstract

Classification of fish is important to assist biologists in environmental monitoring, understanding fish behavior and more. Live fish classification is a challenging problems due to free movements, the light condition and image quality. Recently, deep neural network has shown great performance in image classification and object recognition problems, therefore, transfer learning based on Alexnet is applied on brown trout (Salmo trutta) and European grayling (Thymallus thymallus) images extracted from videos for classification without prior pre-processing. Very high accuracy above 99% and almost perfect F1-score are obtained and this network also can classify the incomplete fish images well with 98% accuracy.