Fish Species Identification in Unconstrained Underwater Environment

Abstract

Preservation of biodiversity is of utmost importance and one of the major ecosystem which is highly in danger and a notable cause of disturbing biodiversity is freshwater ecosystem. Marine researchers and scientists along with government bodies are developing ways to monitor and preserve the underwater living organisms for a sustainable environment. Many techniques have been developed for underwater monitoring and classification but with advancement in knowledge the trends are changing rapidly. An automated system for underwater fish monitoring and classification is needed but the underwater environment is highly variable and complex which poses a great challenge for such automated systems to accurately monitor and identify the inhabitants. We proposed a variant of deep learning method namely Convolutional Neural Network for specie identification in unconstrained underwater environment. We demonstrated how the convolutional layers within a network can provide useful information if used appropriately. The classification was done using dataset obtained from University of Western Australia (UWA). We indicated many reasons that the proposed approach can potentially be used for further classification tasks.