Multi-Feature Sample Database for Enhancing Deep Learning Tasks in Operational Humanitarian Applications

Amongst the many benefits of remote sensing techniques in disaster- or conflict-related applications, timeliness and objectivity may be most critical assets. Recently, increasing sensor quality data availability have shifted attention more towards information extraction process itself. With promising results obtained by deep learning (DL), notion arises that DL is not agnostic to input errors biases introduced, particular sample-scarce situations. The present work seeks understand influence different sample aspects propagating through network layers automated image analysis. In this paper, we broadly discuss conceptualisation such a database an early stage realisation: (1) inherited properties (quality parameters underlying as cloud cover, seasonality, etc.); (2) individual (i.e., per-sample) properties, including a. lineage provenance, b. geometric (size, orientation, shape), c. spectral features (standardized colour code); (3) context-related (arrangement Several hundred samples collected from camp settings were hand-selected annotated with computed initial stage. supervised annotation routine so thousands existing can labelled extended feature set. This should better condition subsequent tasks hybrid AI approach.