Researchers from the Physico-Technical Institute of Petrozavodsk State University together with
For system training two-dimensional imagesdivided into circles with a certain radius, scientists explain. On each of them, the pixels were connected by a line in such a way that the result was a sequence of elements that differ in tone. This approach allows you to convert a two-dimensional image into a one-dimensional series that can be analyzed by the system.
After training the model on several images, itis capable of predicting entropy values with high accuracy already on other images. The effectiveness of the approach was evaluated by the similarity of entropy values predicted by artificial intelligence and calculated by mathematical methods. The study showed that the accuracy reached from 81 to 99%, while the processing time was reduced from eight days to several seconds.
An example of a surface area and the corresponding 2D entropy distribution (top), a cross-sectional plot of the entropy distribution (bottom). Image: Andrey Velichko
Entropy is a measure of chaos, irregularity orheterogeneity, which is used to describe systems of different nature. The ordering of pixels of different "shades" allows us to draw a conclusion about how this or that piece of land is used. Traditionally, special mathematical algorithms are used to process such data.
By analyzing satellite images, one can understandhow much the forest has suffered from fire, strong winds or pest invasion, scientists say. In addition, images taken over the years can be used to gauge how quickly natural landscapes are replaced by cities or agricultural land.
Read more:
Does science exist in extreme conditions? We answer in numbers
Scientists have found out who benefited from the death of dinosaurs
Look at the photo of the Earth from space: it was made by the first private module