A geographic information system for analyzing space data using deep learning neural networks was developed to identify objects of a given geometry. Supercomputer technology and a package of application programs for neural network analysis of Earth remote sensing data have been developed. The system consists of three types of neural networks, database and a tagging system with automatic uploading of the latest satellite images. The developed geoinformation system for marking satellite images has significantly improved the quality of space image recognition. The development is used for automated thematic interpretation of satellite images.