Landsat-8 Sea Ice Classification Using Deep Neural Networks

Knowing the location and type of sea ice is essential for safe navigation route optimization in ice-covered areas. In this study, we developed a deep neural network (DNN) pixel-based Stage Development classification Baltic Sea using Landsat-8 optical satellite imagery to provide up-to-date information Near-Real-Time maritime applications. order train network, labeled regions shown with classes from German Federal Maritime Hydrographic Agency (BSH) charts. These charts are routinely produced distributed by BSH Ice Department. The compiled data set region consists 164 2014 2021 contains types classified Development. level 1 (L1b) images that could be overlaid available based on time acquisition were downloaded United States Geological Survey (USGS) global archive indexed cube better handling. input variables DNN individual spectral bands: aerosol coastal, blue, green, red near-infrared (NIR) out Operational Land Imager (OLI) sensor. bands selected reflectance emission properties ice. output values 4 Free. results obtained show significant improvements compared when moving polygons pixels, preserving original classes. model has an accuracy 87.5% test excluded training validation process. Using can therefore add value safety ice- infested waters high resolution real-time availability. Furthermore, extended other such as Sentinel-2. Our approach promising automated (NRT) services, which deployed integrated at later stage Aerospace Center (DLR) ground station Neustrelitz.