Russian scientists have invented a new method for automatic processing of satellite images. This was reported by the press service of the RNF.
Scientists use the concept of entropy, which has a mathematical expression, to describe the disorder and “chaotic” of objects and phenomena. It is particularly suitable for the analysis of space images to search for irregular areas, for example, man-made objects in a flat area become immediately noticeable. The same applies to natural phenomena: by the arrangement of pixels with different shades of gray, you can distinguish a living forest from a burnt forest suffering from parasites in healthy areas, since living biomass appears more uniform.
The algorithm for this type of analysis was created by experts from the Physical-Technical Institute of Petrozavodsk State University and their German and British colleagues. It calculates the value of entropy not mathematically but based on machine learning: while learning, it processes, classifies, and establishes patterns among a small set of items in a sequence. It can then estimate the entropy values in other images with high accuracy. The effectiveness of the approach was evaluated by the similarity of entropy values estimated by artificial intelligence and calculated by mathematical methods, with high output (accuracy was 81 to 99%).
As a result, the algorithm determined the boundaries of objects with high accuracy, including rivers, forests, roads and fields. Even a dim dirt road was easily detected and showed a wave of entropy. The authors hope that the enhancement will help significantly speed up the processing of space targets for cartographic or environmental purposes.