Comparison of Different Convolutional Neural Network Architectures for Satellite Image Segmentation

Convolutional neural networks for detection geo- objects on the satellite images from DSTL, Landsat-8 and PlanetScope databases were analyzed. Three modification of convolutional neural network architecture for implementing the recognition algorithm was used. Images obtained from the Landsat-8 and PlanetScope satellites are used for estimation of automatic object detection quality. To analyze the accuracy of the object detection algorithm, the selected regions were compared with the areas by previously marked by experts. An important result of the study was the improvement of the detector for the class "Forest". Segmentation of satellite images has found application at urban planning, forest management, climate modelling, etc.