Segmentation of Satellite Images of the Earth's surface using Neural Network Technologies

The authors present an algorithm for segmentation of satellite images based on the U-net convolutional neural network. It includes the following stages: preparation of input data, modification of the structure of a convolutional neural network (CNN) taking into account the parameters of the studied images, training CNN on the formed training sample, and segmentation of test images. The authors suggest procedures for optimizing the CNN training time and increasing the accuracy of the selected classes of objects with a limited training sample of images and computing resources. The results of experimental studies that confirm the effectiveness of the method are presented.