An Effective Algorithm for Analysis and Processing of Satellite Images for Semantic Segmentation

This article proposes an effective algorithm for semantic segmentation of satellite images based on the architecture of the convolutional neural network U-Net. This algorithm constructs raster masks of objects belonging to a single class. This algorithm provides a continuous result without stitches for satellite images of unlimited sizes, despite the limitation of the U-Net input data size. This is achieved by performing the post-processing step. This article discusses the details of the developed algorithm and its application with models trained for segmentation of agricultural fields, forests and urban buildings.