(50-3) 12 * << * >> * Russian * English * Content * All Issues

The accuracy of convolutional neural networks in identifying satellite images of thermokarst polygonal formations
V.V. Zhebsain1, A.Y. Gololobov2, A.F. Poselsky1, N.I. Basharin3

1 North-Eastern Federal University, Belinskogo Str. 59, Yakutsk, 677000, Russia;
2 Yu. G. Shafer Institute of Cosmophysical Research and Aeronomy of Siberian Branch of Russian Academy of Sciences, Prospekt Lenina 31, Yakutsk, 677000, Russia;
3 Melnikov Permafrost Institute of the Siberian Branch of the Russian Academy of Sciences, Merzlotnaya Str. 36, Yakutsk, 677000, Russia

  Full text (PDF)

DOI: 10.18287/COJ1780

Article ID: 1780

Language: English

Abstract:
The paper presents the results of a study of the accuracy metrics for identifying thermokarst landscape changes in space images using neural network technology. The rate of thermokarst degradation of landscapes in regions with permafrost, particularly in Yakutia, has significantly accelerated in recent decades due to global climate warming. The paper discusses the preparation of a training set for neural network numerical experiments. An experimental registry of thermokarst objects has been developed, containing a large number of satellite images of thermokarst polygonal formations. The results of studies of metrics characterizing the accuracy of identification of satellite images of thermokarst formations using a developed applied computer program implementing a multilayer convolutional neural network, the maximum values of which were 0.95-0.96 for Recall and 0.94-0.95 for Precision and F1-score, are presented. 17 series of numerical experiments were performed, consisting of an average of 7-8 experiments, in order to study the dependence of these metrics on convolutional neural models, as a result of which models were selected that provide the best performance in solving problems of identifying thermokarst formations.

Keywords:
convolutional neural networks, machine learning, thermokarst formations, satellite images, training datasets.

Acknowledgements:
This work was financially supported by the Russian Science Foundation under project No. 24-21-20043.

Citation:
Zhebsain VV, Gololobov AY, Poselsky AF, Basharin NI. The accuracy of convolutional neural networks in identifying satellite images of thermokarst polygonal formations. Computer Optics 2026; 50(3): 1780. DOI: 10.18287/COJ1780.

References:

  1. Fedorov AN, Konstantinov PY. Observations of surface dynamics with thermokarst initiation. In: Proceedings of the Eighth International Conference on Permafrost; 2003. pp. 239-243.
  2. Fedorov AN, Konstantinov PY. Observations of surface dynamics with thermokarst initiation. In: Proceedings of the Ninth International Conference on Permafrost; 2008. pp. 433-438.
  3. Konstantinov P, Basharin N, Fedorov A, Iijima Y, Andreeva V, Semenov V, Vasiliev N. Impact of climate change on the ground thermal regime in the lower Lena region, Arctic Central Siberia. Land 2023; 12(1): 19. doi:10.3390/land12010019.
  4. Konstantinov P, Zhelezniak M, Basharin N, Misailov I, Andreeva V. Establishing permafrost thermal monitoring sites in East Siberia. Land 2020; 9(12): 496. doi:10.3390/land9120496.
  5. Witharana C, Udawalpola M, Liljedahl A, Jones M, Jones B, Hasan A, Joshi D, Manos E. Automated detection of retrogressive thaw slumps in the High Arctic using high-resolution satellite imagery. Remote Sens 2022; 14(17): 4132. doi:10.3390/rs14174132.
  6. Runge A, Nitze I, Grosse G. Remote sensing annual dynamics of rapid permafrost thaw disturbances with LandTrendr. Remote Sens Environ 2022; 268: 112752. doi:10.1016/j.rse.2021.112752.
  7. Nitze I, Heidler K, Barth S, Grosse G. Developing and testing a deep learning approach for mapping retrogressive thaw slumps. Remote Sens 2021; 13(21): 4294. doi:10.3390/rs13214294.
  8. Huang L, Liu L, Luo J, Lin Z, Niu F. Automatically quantifying evolution of retrogressive thaw slumps in Beiluhe (Tibetan Plateau) from multi-temporal CubeSat images. Int J Appl Earth Obs Geoinf 2021; 102: 102399. doi:10.1016/j.jag.2021.102399.
  9. Yin G, Luo J, Niu F, Lin Z, Liu M. Machine learning-based thermokarst landslide susceptibility modeling across the permafrost region on the Qinghai-Tibet Plateau. Landslides 2021; 18(12): 3939-3955. doi:10.1007/s10346-021-01669-7.
  10. Lara HA, Bouchard F, Saint-Jean G, Farid G, Laurent E. Automated identification of thermokarst lakes using machine learning in the ice-rich permafrost landscape of central Yakutia (Eastern Siberia). Remote Sens 2023; 15(5): 1226. doi:10.3390/rs15051226.
  11. Zhebsain VV, Gololobov AYu, Basharin NI, Poselsky AF. Features of the application of neural network technology for the recognition of graphic images of thermokarst objects [in Russian]. Geoinformatika 2024; 4: 39-47.
  12. Bagaeva II. Analysis of the concepts of neural network and convolutional neural network. Training a convolutional neural network using the TensorFlow module [in Russian]. Matematicheskoe Modelirovanie i Programmnoe Obespechenie Sistem v Promyshlennoy i Sotsialnoy Sferakh 2020; 8(1): 15-22. doi:10.18503/2306-2053-2020-8-1-15-22.

151, Molodogvardeiskaya str., Samara, 443001, Russia; E-mail: journal@computeroptics.ru; Tel: +7 (846) 242-41-24 (Executive secretary), +7 (846) 332-56-22 (Issuing editor), Fax: +7 (846) 332-56-20