Roscosmos State Corporation seeks to provide these to consumers in the highest quality and most suitable for use. To achieve this goal, the key companies of the State Corporation are developing special software. It is embedded in the data processing process and improves the quality of the output. When creating such solutions, neural network technologies are increasingly used. Developments are conducted by specialists of the Russian Space Systems Companies, Research Institute of TP and Terra (everyone is part of Roscosmos).
Developed neural network algorithms are integrated into the information system “Digital Earth”. Algorithms solve two classes of tasks. The first one is to improve the quality of the pictures. For him, the Super Resolution, Neural Network Panshaprening algorithms, smoke compensation algorithm and correlation algorithms have been developed. The second class of solved tasks is the creation of a quality mask where the algorithms segmentation of clouds, shadows, hasshes, defects of the image and classes of the underlying surface are used.
For learning neural networks, the created and constantly updated standards database is used. More than a year, specialists of state corporation companies are working on thematic marking and relevant data preparation. Currently, more than 4 million square kilometers of data were processed and more than 160 thousand standards were prepared for various objects.