The Application of Deep Learning in Marine Sciences

Abstract

Ecological studies are increasingly using video image data to study the distribution and behaviour of organisms. Particularly in marine sciences cameras are utilised to access underwater environments. Up till now image data has been processed by human observers which is costly and often represents repetitive mundane work. Deep learning techniques that can automatically classify objects can increase the speed and the amounts of data that can be processed. This ultimately will make image processing in ecological studies more cost effective, allowing studies to invest in larger, more robust sampling designs. As such, deep learning will be a game changer for ecological research helping to improve the quality and quantity of the data that can be collected. Within this chapter we introduce two case studies to demonstrate the application of deep learning techniques in marine ecological studies. The first example demonstrates the use of deep learning in the detection and classification of an important underwater ecosystem in the Mediterranean (Posidonia oceanica seagrass meadows), the other showcases the automatic identification of several jellyfish species in coastal areas. Both applications showed high levels of accuracy in the detection and identification of the study organisms, which represents encouraging results for the applicability of these methodologies in marine ecological studies. Despite its potential, deep learning has yet not been widely adopted in ecological studies. Information technologists and natural scientists alike need to more actively collaborate to move forward in this field of science. Cost-effective data collection solutions are desperately needed in a time when large amounts of data are required to detect and adapt to global environmental change.