Deep Learning Framework for Waveband-Specific Global Scene Background Generation

The proposed effort aims to develop a deep learning framework for waveband-specific and geospecific background generation to address US Air Force’s need to create a global, multispectral background scene database to support scene generation. Using deep neural networks and collections of Geographic information system (GIS) data, we propose to develop a tool for background database generation that automates the extraction scene background features within a user-defined Area of Interest (AOI) on Earth, leverages this information to resolve imagery at the apparent resolution of sensors rapidly changing altitudes, and streamlines the preparation of generated backgrounds scenes for processing in FLITES. The proposed solution requires the development of tools to interface with GIS databases for extraction of available vector and raster data; semantic segmentation models to classify land cover into material classes with known physical properties; and texture synthesis and super-resolution models to enhance imagery, simulating reductions in ground sample distance. The Phase II effort will focus on capability extension of algorithms developed in Phase I, large-scale GIS data curation, optimization of algorithms for accuracy and efficiency, maturation of the software tool to enable compatibility and integrability with FLITES, extensive verification and validation, and technology insertion into Air Force workflows.