Terrain Segmentation Using a U-Net for Improved Relief Shading

Since landforms composing land surface vary in their properties and appearance, shaded reliefs also present different visual impression of the terrain. In this work, we adapt a U-Net so that it can recognize selection segment We test efficiency 10 separate models apply an ensemble approach, where all are combined to potentially outperform single models. Our algorithm works particularly well for block mountains, Prealps, valleys, hills, delivering average precision f1 values above 60%. Segmenting plateaus folded mountains is more challenging, rather scattered due smaller areas available training. Mountains formed by erosion processes least recognized landform because similarities with other landforms. The highest accuracy one 65%, while 61%. relief shading techniques were found be efficient regarding specific within corresponding segmented blend them together. Finally, trained model best on mountainous around world, proves work regions beyond training area.