Project

REVO: REvealing the Voices of the Ocean

Motivation

While the North Sea is under significant anthropogenic pressure (from shipping, offshore construction, coastal development, overfishing, and climate change), monitoring the impact on the underwater ecosystems remains challenging. The current common approach involves visual surveys through diving. However, this method has limitations, particularly in the North Sea, where diving is restricted in some areas, and visibility is low due to the high turbidity of the water. By using new technologies, data from various sensors could replace diving in such areas, providing invaluable information about the state of ocean biodiversity.

Within the NLAS project (Next Level Animal Sciences) the Animal Science Group is developing a smart biodiversity sensing box. Currently it is a static box that combines the collection of:

  1. Video recording;
  2. Water sampling for environmental DNA (eDNA) analysis; and
  3. Underwater sound recording.

This box already enables to gather information about the underwater biodiversity in place where diving is prohibited or dangerous. In the future, the Biodiversity box will be smarter, meaning that the eDNA sampling will be triggered based on real time video and sound on board analysis revealing the presence of unknown species or specific species of interest.

Aim & Objectives

The end goal of this EngD project is to trigger eDNA sampling based on real-time sound event detection. Several components are necessary to achieve this goal:

  • A high-resolution acoustic machine learning model for marine animal sounds detection and identification. While this has been done for terrestrial species and marine mammals, it is still in its infancy for other marine animals.
  • An online database to visualize, collect, store and share sounds of interest from the North Sea. This will be developed in collaboration with stakeholders interested in underwater acoustics in the North Sea.
  • Improving the hardware of the current box to transform it into an autonomous embedded system capable of acquiring, storing, processing and communicating data.
  • A system to trigger eDNA sampling based on a relevant situation indicated by on-board acoustic monitoring interpretation.
  • Subsequently, a dashboard combining sounds, video, and eDNA will be developed for researchers, stakeholders, and the interested public to observe in real-time the marine animals present in locations not easily accessible to them.

Methodology

Building on existing knowledge of machine learning algorithms applied to bird and marine mammal recognition, several methods will be tested to develop a robust detection and classification algorithm for other marine animals. Audio recordings will be transformed into spectrograms (Fig. 1), and the AI algorithm will be trained to recognize sounds of different species.

Figure 1: Haddock's Hum. Source: https://fishsounds.net
Figure 1: Haddock's Hum. Source: https://fishsounds.net

There is currently a significant gap in knowledge regarding fish sounds and an even greater one in invertebrate sounds, making it challenging to identify species based on their sounds. Over time, the algorithm will include the detection and identification of more species as knowledge about marine animal sounds progresses.

The algorithm will then be integrated into the embedded computer of the box and the dashboard (Fig 2.) will be developed.

Figure 2: Deployment and functions of the Biodiversity Sensing Box.
Figure 2: Deployment and functions of the Biodiversity Sensing Box.

Feel free to contact me if there is an opportunity to collaborate, if you are interested in sharing acoustic recordings of the North Sea or if you would like to participate to this project as a student. I would be glad to answer questions about my project.

Media

Click here to hear the hum of a Melanogrammus aeglefinus