Search academic texts

Information × Registration Number 0221U106732, 0121U114416 , R & D reports Title Deep learning methods and models for applied problems of satellite monitoring popup.stage_title Head Kussul Nataliia M., Доктор технічних наук Registration Date 18-12-2021 Organization Space Research Institute of National Academy od Science and State Space Agency of Ukraine popup.description2 Purpose of work - improvement of existing and development of new deep learning methods and models to solve problems of monitoring economic indicators and business intelligence based on the fusion of heterogeneous satellite data and data from different sources of high spatial and temporal resolution. Research methods - machine learning methods, statistical analysis, information technologies of big data analysis, improvement of deep learning models, use of recurrent neural networks, method of increasing spatial distinction of satellite data based on deep learning methods, methods of mathematical statistics and geospatial analysis. Using cloud environments for data processing. A method for adaptation recurrent neural networks to integrate and harmonize satellite data. Results: The architecture of the recurrent neural network with the output in the form of a multidimensional time series is implemented, the general scheme of its software implementation and testing procedure are described. The improved architecture of the model of deep learning for the classification of time series based on transfer learning is implemented, the general scheme of its implementation and testing is described. Procedures for testing the results of a recurrent neural network with output in the form of a multidimensional time series and network results based on transfer learning are developed, the general scheme of their implementation is described. The offered method of increase of spatial distinction of satellite data on the basis of deep learning methods with use of Generative Adversarial Networks is realized, the scheme of program realization is described. The proposed methods of merging satellite data of high and low spatial resolution are implemented; the scheme of software implementation is described with justification of the choice of means of implementation. The procedure of testing the results of the method of increasing spatial resolution. Product Description popup.authors Yemelianov Mykhailo O. Zhukovskyi Anton L. Krasilnikova Tetyana M Lavreniuk Mykola S. Parhomchuk Oleksandr M. Yailymov Bohdan Ya. popup.nrat_date 2021-12-18 Close