A brand new European satellite will utilize machine learning to provide quick, inexpensive information on soil conditions, enabling better farming. This project serves as a model of what new sensors and artificial intelligence technology could achieve in a smaller car than an empty shoebox.
Edge computing can be described as a trendy buzzword for shifting processing power away from server farms on the internet and to the places where data is being processed. According to certain people, edge computing is the future of technology. For satellites, in which bandwidth for communications is highly restricted, it could transform the way we communicate.
This European Space Agency’s Intuition-1 satellite program will offer soil information to help drive European Precision Agriculture initiatives that will apply fertilizer when and where necessary, rather than treating the entire field. Precision agriculture is also cheaper and less harmful to the environment. The problem is it needs a lot of details about the soil’s conditions on a tiny scale. The current method of determining the soil nutrients with sufficient precision involves collecting samples from several sites and sending them to a lab to be analyzed. It typically takes three weeks.
The Intuition-1 satellite is a hybrid of hyperspectral sensing and machine-learning in a compact package.
Intuition-1 views the world using the eyes of a hyperspectral sensor. The camera can view every patch of soil with several different wavelengths at once, effectively allowing for a larger spectrum of colors than our human vision. Through comparing the image with different wavelengths, researchers can determine their chemical structure.” It could be possible to measure specific parameters like the amount of magnesium, potassium phosphorus, or the pH of the soil,” states Zbigniew Kawalec, the director of QZ Solutions, the IT company located in Opole, Poland, which is currently developing the analysis part.
Usually, such analyses are performed via a server located in the ground. However, since hyperspectral imaging is massive amounts of data for a single image of a few thousand hectares takes up gigabytes which is enough data to transmit to Earth. The best solution is to execute the processing of hyperspectral analysis by using machine-learning algorithms that are efficient. All that the satellite will need to transmit is its results regarding soil composition.
KP Labs, Polish experts of autonomous systems for space-related applications, have created and tested a working prototype with high-resolution images from fields from Southern Poland captured by aircraft. The technology can be utilized for aerial photography, but the researchers claim that satellites provide more excellent coverage at a lower cost. The next phase will confirm that the method works with hyperspectral images gathered from satellites.
Intuition-1 can be described as a 6U or six-unit Cubesat. It is built on six ten-centimeter cubic units (3x2x1). It will cover the entire surface with a pixel size of 30 meters. The soil information will become available within a few days instead of weeks and without the hassle of continuously taking soil samples and processing them. The satellite is set to launch equipped with a hyperspectral camera, machine-learning algorithms towards the close of 2022.
The project is financed by the Ph-Lab mission of the ESA (pronounced Philab) mission to discover new methods to use satellites for earth observation. The research builds upon the findings of an earlier Ph-lab research project that utilized neural networks to determine the amount of moisture of soil and sea ice which demonstrated that this type of processing in orbit is an effective method.
Intuition-1 demonstrates how much edge computing power can be condensed in tiny packages. A brand next generation of intelligent satellites utilizing similar technology for different types of earth-sensing are likely to follow. The same approach to machine learning is already being used in other applications for small robots that range from drones for mass-market use to underwater vehicles that aren’t human-crewed, which are used to locate underwater mines and robot dogs exploring underground tunnels. In areas where communication is not available, edge computing using machine learning can handle processing locally. Intuition-1 proves that the sky’s not the limit.