DETECTION OF CLOUDS IN MEDIUM-RESOLUTION SATELLITE IMAGERY USING DEEP CONVOLUTIONAL NEURAL NETS

Commercial platforms like WorldView-3 present a unique source of data for many applications due to their sub-meter resolution. To make the data usable for land feature detections, automated cloud screening is needed. WorldView-3 currently reports a single cloudy percent number for the scene, often inaccurate. The method is proprietary. We show a new general-purpose cloud masking method for WorldView-3 that is free to use and develop further. We show a variety of results and initial validation.