Cloud cover detection using a neural network based on MSU-GS instrument data of Arktika-M No 1 satellite
V.D. Bloshchinskiy1,2, L.S. Kramareva1,2, Yu.A. Shamilova2
1Computing Centre Far Easten Branch of the Russian Academy of Science, Khabarovsk, Russia
2Far-Eastern Center of State Research Center for Space Hydrometeorology В«Planeta», Khabarovsk, Russia
Keywords: MSU-GS, Arktika-M, cloud mask, cloud detection, neural network classifier, U-Net
Abstract
The paper presents an algorithm based on a convolutional neural network with a modified U-Net architecture for detecting cloud formations in satellite images. Multispectral satellite images obtained from the MSU-GS instrument installed at Arktika-M No 1 satellite are used as input data. The accuracy of the algorithm was evaluated using machine learning metrics and comparing the results with reference masks compiled by manual decryption of the satellite images by an experienced decoder specialist. In addition, a comparison with similar products based on data of the SEVIRI and VIIRS instruments was conducted. For areas illuminated by the sun, the cloud mask obtained by the proposed algorithm has an accuracy of 92% compared to the reference mask, and for areas not illuminated by the sun, the accuracy is 89%.
|
