Coding in the deep: the past, present & future of deep-learning in the field of benthic ecology

Written by Rhodri Irranca-Davies

Around the world there is now a growing demand for more comprehensive monitoring of Earth’s vulnerable deep-sea habitats and species. Not only will this help us to better understand the myriad of anthropogenic effects that we as humans are subjecting on the planet, but also protect the diverse spectrum of animals that reside on or around the seafloor. Traditionally scientific research on deep-sea environments has mainly constituted invasive methods, such as benthic trawling and sediment coring. These techniques cause widespread damage to marine species by impacting them in both direct ways (damaging or removing individuals) and indirect ways (changing their environment). These species are often long-lived and slow to mature, compared to their shallow-water counterparts. Therefore they often have slower recovery rates, sometimes taking decades or longer to fully recover from these impacts.

identification
An example image taken using benthic imaging technology that identifies 12 different species in just one small area

Benthic imaging technologies

Luckily less-destructive alternatives are now allowing for deep-sea monitoring without the disturbance to benthic habitats and species.These mainly comprise of the use of specialised deep-sea imaging technologies capable of capturing high quality photos and video footage. They can be attached to remote operated vehicles (ROVs), automated underwater vehicles (AUVs), and towed video sleds. All of which are part of an industry that has undergone explosive advancement since the mid-1900s, due to a reduction in price because of better underwater housing, image quality, batteries and data storage. In return researchers can learn more about deep-sea ecology than just presence and density of species, they can also monitor complex things like animal behaviour, species interactions, ecosystem functions and resilience to change, as well as recovery rates.

timeline

The data processing bottleneck

These technological advancements have made sampling the deep-sea relatively cheap and much more accessible and repeatable. This means that we are now able to collate data across larger spatial (metres to kilometres) and temporal scales (hours to years), leading to an exponential increase in data collection. However this has generated a new problem for those scientific professionals – manually processing the data. Which in turn has led to the formation of a data processing bottleneck. Ultimately what this means is that there is an inability to analyse the data at a rate equivalent to or greater than that at which it is collected and stored.

bottleneck

Deep-learning

As a result of this data processing bottleneck there is accelerating focus within benthic ecology on utilising computers to faster process data. To do this researchers are started to rely on a combination of artificial intelligence and computer vision known as deep-learning. This involves the application of multiple layers of machine learning algorithms, known as neural networks, to achieve improved outputs from the raw imagery. Essentially using computers to analyse photos and videos to identify, count and describe marine species. These techniques have already achieved formidable results in various tasks, often directly competing with or surpassing the results of human analysts. As the technology continues to improve it will allow researchers to completely bypass the bottleneck, and so be able to drastically increase the monitoring of deep-sea ecosystems.

deep learning

Further applications

The deep-learning capabilities of computers have applications in sectors far removed from ecology. Broadly speaking, they can be used for the detection of any semantic object, from animals caught in camera traps to fruit hanging in a tree. Moreover it has growing application in facial recognition software and the visual tracking of individuals through crowds for use by security, defence, and military services globally. The applications that deep neural networks currently have, and could come to have, in healthcare and medical imaging are boundless. From identifying clusters of cancer cells from cardiograms to applications in autonomous in-home pre-natal health checks for pregnant patients. However for now this technological breakthrough is hard at work exploring the deepest parts of our oceans.

Rhodri Irranca-Davies is a marine biology student currently undertaking his MRes at Plymouth University, where he studies growing applications of artificial intelligence and computer vision in deep-sea benthic ecology. He has also completed his BSc Applied Marine Biology at Bangor University, with a placement year with the Specialist Marine Monitoring Team at Natural Resources Wales. You can follow him on Twitter and Instagram or via his LinkedIn page here.

If you have an ocean story to tell then Marine Madness is here to help you tell it. If you would like to contribute check out the submission guidelines here.


Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s