An early attempt to identify seals from aerial images of the sea ice.


Detection of sea mammals from aerial imagery

Efficient and reliable methods for automatic analysis of complex marine observation data are needed to ensure sustainable fisheries and harvest. The Institute for Marine Research (IMR) has a mission to be a leader in providing knowledge to ensure sustainable management of resources in our marine ecosystems.

Abundance estimation of sea mammals like ice breeding seals (harp and hooded) and coastal seals (grey and harbour) are based on aerial photographic surveys. The Institute of Marine Research (IMR) utilizes aerial images of the sea ice captured from above as a data source to estimate abundance of seal populations.

An aerial image cropped down to an identified seal pup.

To achieve this in a reliable and cost-effective manner better automation of this process is needed and will lead to better long-term management of seal populations. Previous activities have demonstrated that there is a potential in using deep learning for this. However, limitations in training data due to incomplete annotations and shifts in sensors and conditions over time are challenges that need to be solved.

For the detection and classification of seals on ice, the datasets will come from both drones and manned aerial surveys. We will start from previous models and extend these to include more species (coastal seal species). Furthermore, we need to improve on these models to better handle incomplete annotations and shifts in sensors and conditions over time, with the aim of limiting false positives, and to account for large fractions of images without seals.


Further reading

Performing objective measurements in ultrasound images
March 8, 2021
Exploiting limited data to perform objective measurements in ultrasound images of the heart.
Oil-spill detection and characterization of thickness
December 18, 2020
Visual Intelligence collaborates with KSAT to develop new models for detecting and characterizing oil spills.
Deep learning and AI in the medical domain
January 19, 2021
Overcoming the challenges of limited training data in the medical domain and laying the fundamentals for explainability and reliability.