Marine sciences

Developing models and applications to monitor the marine environment


Ecological studies, involving e.g. classification and statistical counting of species in an ecosystem, has been challenging and time consuming tasks in the marine sciences. Efficient and reliable data-driven methods for automatic analysis of complex marine observation data are needed to ensure sustainable fisheries and harvest. Deep learning has the potential to automate and streamline the steps required in such studies, but few applications in this domain has been developed.

An important source of data in marine sciences is the echo sounder which is used to observe the marine ecosystem at a larger scale. Some preliminary approaches using convolutional neural networks for fish detection and classification from acoustic data are emerging but very little research has been done in this area.

Main objective

The innovations in the field of marine sciences aim developing  efficient and reliable deep learning methods for automatic analysis of complex marine observation data.


Many of the challenges related to the use of deep learning in this area are related to training data, where the amount of annotations can be limited and the quality variable and it is too expensive to get more and/or better data. There is also a need for explainability and reliability, as trust becomes very important when the output from these systems are intended as input to models for abundance estimation which again is a basis for setting of fishing quotas.

Visual Intelligence is advancing deep learning in the marine sciences to overcome these challenges.

Highlighted publications

On the Effects of Self-supervision and Contrastive Alignment in Deep Multi-view Clustering
December 19, 2023
We propose DeepMVC – a unified framework which includes many recent methods as instances.
Merging clustering into deep supervised neural network
June 8, 2023
Introducing the SuperCM technique to significantly improve classification results across various types of image data.
Addressing Distribution Shifts in Federated Learning for Enhanced Generalization Performance
June 4, 2023
Training and test data from different clients pose a challenge.