On some approaches to machine learning of convolutional neural networks in the tasks of processing aerial images of various presentation formats

The presented scientific paper presents one of the approaches to solving the problem of pattern recognition on an array of initial data of aerial photographs of various resolutions and presentation formats. Based on a critical analysis of the existing tools, a convolutional neural network mechanism was chosen to solve the problems of classification, detection, segmentation and analysis of changes with a small discreteness of changes. The author of the paper proposed a software implementation of the selected recognition method, including solving problems of modeling real-world objects on a cartographic basis.