A neural network will observe the Baikal plankton


Since February 1945, employees of Irkutsk State University lead continuous observations of the plankton community of Lake Baikal. Opposite the biological station of the Research Institute of Biology of Irkutsk State University in the village of Bolshiye Koty, at a distance of 2.7 km from the coast, there is the so-called “Point No. 1”, where weekly samples of phyto- and zooplankton are taken and then analyzed from depths from 0 to 800 meters. In the early years, these studies were conducted to assess the fish productivity of the lake. For example, the diet of the Baikal omul is based on zooplankton, which, in turn, feeds on phytoplankton. Therefore, the size of the fish population depends on the species composition and number of microscopic algae. Now, monitoring of the lake’s plankton is needed primarily to assess the ecological state of Lake Baikal itself.

“The phyto- and zooplankton community is essentially the foundation of the entire ecosystem of Lake Baikal. Understanding the processes in this foundation, their dynamics, we can make forecasts on the vectors of development of the entire ecosystem of the lake. Monitoring project “Point No. 1” is unique in that it allows you to make an analysis based on long-term and continuous series of observations accumulated over 75 years, – tells Maxim Timofeev, Doctor of Biological Sciences, Director of the Research Institute of Biology, ISU. – The greatest time and labor costs are spent in microscopy and recognition of species and forms of plankton in samples. This can only be done by a specialist. And not a simple biologist, but a specialist in the groups analyzed in the sample. Some of them specialize in phytoplankton, others in zooplankton. The training of such a specialist, who is able to distinguish, for example, 400 forms of phytoplankton, takes years. “.

Because of this, in recent years, the project “Point # 1” was under threat of closure – the data recognition method, which is now used in the project, is technologically outdated. To maintain the project, several dozen high-level specialists would be required, agreeing at the same time to perform routine operations. In order to preserve and continue to develop the unique project, the joint team of scientists and developers is creating a neural network algorithm that will automatically recognize and classify microorganisms in samples of Baikal water.

To train the algorithm, scientists from the Research Institute of Biology of Irkutsk State University provided more than 1000 images of each species of Baikal plankton to developers of artificial intelligence models for studying marine ecosystems MaritimeAI… Based on these data, a classification mechanism for plankton species will be created using Yandex DataSphere – service Yandex.Cloud for data analysis, development and operation of machine learning models. Images of microorganisms will be transmitted to Yandex.Cloud directly from the microscopes of the laboratory of the Research Institute of Biology, ISU. It is assumed that the algorithm will be able to detect up to 99% of all types of plankton, and specialists from the Institute of Biology will control the quality of its work.

“We train algorithms using a fairly standard gradient descent algorithm with variations in this area, but we pay more attention to the architecture of neural networks. Colleagues from the research institute collect images of objects from under a microscope, and we begin training neural networks from the already collected array of images. Of course, no one can talk about an array of images for 75 years of observations, because the ability to save an image from a microscope to a computer appeared not so long ago, but one of the important points in our work was changing the analysis process so that in the future the accumulation of images would be its integral integral part “, – comments Sergey Oreshkov of MaritimeAI.

Although now there are many online projects to attract scientific volunteers to process large data sets, for example, for classification astronomical images or definitions of animals on video recordings, in the task of recognizing Baikal microorganisms in samples, it is the experience of specialists that is important, to help whom a new neurointerface is being created. At the same time, the help of volunteers can be useful in the process of creating the neural network itself. To train algorithms, it is important to provide them with high-quality labeled data – a task that could be done by volunteers of science. For them, specialists MaritimeAI are preparing the interfaces of the Yandex.Toloki project. Researchers and developers promise to present a working prototype of the neural network this summer.

According to the press release Yandex.


Source: Автономная некоммерческая организация "Редакция журнала «Наука и жизнь»" by www.nkj.ru.

*The article has been translated based on the content of Автономная некоммерческая организация "Редакция журнала «Наука и жизнь»" by www.nkj.ru. If there is any problem regarding the content, copyright, please leave a report below the article. We will try to process as quickly as possible to protect the rights of the author. Thank you very much!

*We just want readers to access information more quickly and easily with other multilingual content, instead of information only available in a certain language.

*We always respect the copyright of the content of the author and always include the original link of the source article.If the author disagrees, just leave the report below the article, the article will be edited or deleted at the request of the author. Thanks very much! Best regards!