A method of neural network identification of thermokarst formations in Central Yakutia using graphic data from satellite remote

Abstract This paper presents a method for identifying thermokarst formations in Central Yakutia using remote sensing satellite images based on neural network technology. The method and methodology for generating data sets for training and testing neural networks containing four classes of landscape types, such as landscapes containing thermokarst formations, fields, forests and water bodies, are described. The structure and mathematical description of the convolutional neural networks forming the basis of the method for identifying thermokarst features are presented. The results of numerical experiments on the study of the accuracy of identification of satellite images of thermokarst formations and other landscape classes using the developed applied computer program are presented. Based on an analysis of the results of numerical experiments, three convolutional neural network models were selected that yield the best results in identifying thermokarst features, with accuracy levels of approximately 0.94–0.96.