Sea Ice Cover Detection of the Far Eastern Seas by Data of the MSU-MR Radiometer of the Meteor-M No. 2 Satellite

In this paper, the technology for determining the ice cover mask using a convolutional neural network as applied to data of a low-resolution multispectral scanning device installed on the Meteor-M No. 2 Russian satellite is considered. The selection criteria for the parameters involved in training the neural network and the process of determining texture size are described. The classification score of the developed model is determined using the machine learning metrics. Validation of the results shows that the algorithm has an accuracy of 94.9 and 96.7% in comparison with ice cover masks according to data of the MOD10 product of the MODIS instrument and archived ice condition maps created in accordance with the international WMO Sea Ice Nomenclature.