LUT taught AI to recognise marine plankton
Phytoplankton releases oxygen into the Earth's atmosphere and is a significant element in the food chain of marine organisms. LUT University's computer vision and pattern recognition expertise is utilised in plankton research.
Automated computer recognition is a rapid method for identifying tiny plankton cells, which have previously been identified with the human eye through a microscope lens.
"From a hundred image samples taken from the Baltic Sea, we have taught artificial intelligence to recognise over 34 types of phytoplankton with 84 per cent certainty. We need to have enough images to achieve a high recognition rate," explains Heikki Kälviäinen, Professor of Computer Science and Engineering.
The image data from the Utö island in the Baltic Sea is produced by the Finnish Environment Institute. High-quality flow cytometers* capture an image of filtered sea water each time an object resembling plankton passes the lens.
"The quantity of data is very large and its manual processing is extremely laborious. Training an automated species recognition system takes weeks and a great deal of computational capacity, but once it's done, the recognition software is fast to use," Kälviäinen says.
By recognising plankton species, marine biologists in the Finnish Environment Institute are better able to understand the composition and ecology of the plankton community with a higher time resolution. The data also provides information on the state of the Baltic Sea and the global ecosystem.
"The amount and community composition of phytoplankton reflects, for instance, the impacts of eutrophication and climate change on water quality, as plankton species react quickly to environmental changes. Alterations in the composition of the plankton community anticipate changes at higher levels of the food web," analyses Senior Researcher Sanna Suikkanen from the Marine Research Centre of the Finnish Environment Institute.
Also the identification of harmful and invasive plankton species requires knowledge of plankton communities.
Identification tool roaming the seas
LUT started developing a plankton recognition method in 2018. Osku Grönberg wrote his Master's thesis on the topic: Plankton recognition from imaging flow cytometer data using convolutional neural networks.
The objective for the future is that plankton cells be recognised as rapidly as possible after taking the image. LUT, in collaboration with the Finnish Environment Institute, is developing an applicable method in the FASTVISION project, which is funded by the Academy of Finland and was launched at the beginning of September 2019.
"Data produced by automated imaging hardware in the sea is fed to species recognition software," Kälviäinen describes the method.
The project examines the composition of phytoplankton in the Baltic Sea and its seasonal and annual changes. The data is compared to the marine monitoring data collected by the Finnish Environment Institute.
"The data shows the range of species in the Baltic Sea in specific locations in the summer, winter, or a decade ago," Kälviäinen elaborates.
In addition to regional data, the FASTVISION project is anticipated to reveal whether changes in the Baltic Sea plankton community could be generalisable to other European marine areas.
The project is challenging but not impossible. Kälviäinen is especially fascinated by the fact that LUT could respond to the need for digital expertise in nature conservation. He also expects to make observations on transfer learning in neural networks. Deep neural networks are a commonly used tool with applications in all areas of computer vision and artificial intelligence.
"It will be interesting to see whether the species recognition software, which has been taught to recognise the species community in one area, could be adapted to a new environment and species community with a minimal number of sample images," Kälviäinen says.
*Flow cytometry is a technique used to calculate, group or otherwise define cells or other particles in a fluid by injecting the cells individually through a small hole or pipe into the cytometer. Source: Helsinki term bank for the arts and sciences
FASTVISION – life in the fast lane
- The Academy of Finland has granted two-year financing to the FASTVISION project, which started in September 2019.
- The project brings together the Finnish Environment Institute's plankton imaging software and taxonomic expertise and the LUT Computer Vision and Pattern Recognition Laboratory's (CVPRL) expertise.
- LUT will develop plankton recognition software programmes by applying computer learning methods. The technique will recognise plankton species more quickly and efficiently than the human eye.
- Plankton is food to many marine organisms. It plays a significant role in global and marine ecosystems.
- The Baltic Sea and its organisms are threatened by global warming and eutrophication.
Heikki Kälviäinen, Professor of Computer Science and Engineering, Project Director, LUT University
tel. +358 40 586 7552
Sanna Suikkanen, Senior Researcher, Finnish Environment Institute tel. +358 295 251660
- Computer vision identifies the endangered Saimaa ringed seal – enables protection of the species
- Computer vision enables the analysis of nanoparticles in industrial and medical applications