Scientists from Skoltech have created a machine learning system that will help space agencies of the world to select the "right" plants to provide future long-term space missions with the required amount of biomass and oxygen. Their findings were presented in the journal IEEE Pervasive Computing.
“The main advantage of our method is that it is enough to obtain a three-dimensional image for each plant species only once. After that, to predict the biomass growth, it is enough to use the simplest cameras. This greatly simplifies and reduces the cost of forecasting, control and optimization systems for greenhouses and artificial life support systems in space,”notes Dmitry Shadrin, a Skoltech graduate student, quoted by the university's press service.
Long-term space flights, according to NASA and Roscosmos experts today, will require the creation of fully autonomous life support systems that allow the production of water, oxygen and all the necessary nutrients for an unlimited time.
Plants and various unicellular algae, capable of producing biomass in large quantities and at high speed, are considered the key to their creation today. Over the past two decades, scientists have made significant progress in this direction, creating two greenhouses aboard the ISS and growing cabbage, lettuce, asters and many other plants in them.
Such successes make biologists, space doctors and other researchers wonder how many plants are needed for the survival of a crew flying to Mars or other planets. Their excess can make the mission too expensive and unrealizable, and the lack - doom future followers of Mark Watney from "The Martian" to a slow death.
Despite the fact that scientists have been studying plants for thousands of years, it is not so easy to prepare such estimates, since the rate of their growth and biomass gain depends on many different biological and physical factors - the amount of moisture and trace elements in the soil, the level of illumination and dozens of other things. In addition, the biomass itself is rather difficult to "weigh" without killing the plant itself, which interferes with the assessment of its growth rate.
Shadrin and his Skoltech colleagues, Rupert Gerzer, Tatiana Podladchikova and Andrey Somov, figured out how to quickly and accurately make such assessments by observing the growth of dwarf tomatoes using 3D and 2D cameras.
By analyzing the state of tomatoes at different growth phases, Russian scientists were able to deduce several patterns associated with the biomass set and used them to create machine learning systems capable of assessing these characteristics by analyzing simple two-dimensional photographs of tomato leaves and a three-dimensional model of the plant.
Promotional video:
Further observations showed that this program correctly predicted the growth rate of tomatoes, as well as several varieties of lettuce, during the first 30 days of their life after planting. This allows it to be used not only for calculating "space" life support systems, but also for optimizing the operation of greenhouses.