The machine learning algorithm (ML) analyzes water samples from the lake, identifies and classifies the microorganisms contained in it. Already, the neural network is able to work with 70 forms of plankton, which are most often found in samples. The introduction of artificial intelligence will simplify the work of biologists who have been counting and identifying microorganisms manually for many years.
The neural network in the project helps biologists automate the entire monitoring cycle and, subsequently, get data for new research faster. Together with the scientists of the Research Institute, the neural network was developed by the MaritimeAI company, the Yandex Cloud platform team and the Lake Baikal Foundation for Applied Environmental Research and Development.
Biologists provided almost 50 thousand images of samples, of which 20 thousand were used to train algorithms. Now the images of samples from microscopes are automatically transferred to the Yandex Cloud Cloud platform. The algorithm determines the smallest crustaceans, their species and generates report cards. The neural network continues to be trained in the Yandex DataSphere ML algorithms development and operation service. The data was marked up using the Toloka crowdsourcing service.
"The scientific community and educational organizations are making more and more discoveries in the cloud. One of the priorities of our platform is to create a reliable springboard for easy use of cloud services in research projects. Yandex Cloud launched a crop monitoring system, created an algorithm for an unmanned racing car, and investigated dark matter. The neural network for ecological monitoring of Lake Baikal is a special project for us and for the whole community, incredible in its scale and significance."
Alexey Bashkeev
General manager
In the future, the project participants plan to scale up monitoring and monitor the state of water in other points of Lake Baikal. Also, developers will consistently upload technologies that are used in the project to opensource. So, a dataset of sample images with markup is already freely available. It can be used to test hypotheses on the detection, segmentation and classification of planktonic organisms. A more complete data set and source code will be posted later. This will help other scientific groups and institutes around the world to develop their own water monitoring systems.
