Animal habitat interpretation with KERAS library deep learning methods

DOI: 10.35595/2414-9179-2025-2-31-54-65

View or download the article (Rus)

About the Authors

Andrey V. Korosov

Petrozavodsk state University,
33, Lenin Ave., Petrozavodsk, 185910, Russia,
E-mail: korosov@psu.karelia.ru

Natalia A. Marfitsyna

Petrozavodsk state University,
33, Lenin Ave., Petrozavodsk, 185910, Russia,
E-mail: marfitsyna.nata@mail.ru

Abstract

This work focuses on the use of neural network-based approaches for the purposes of classifying clear-cut areas of different ages using remote sensing methods. This study is based on the field survey results conducted in the territory and medium spatial resolution images from the Landsat-8 satellite. Unlike the reference decoding technology, the classification of secondary forest types (at the site of clear-cut areas of different ages) was assigned in the procedure for joint classification of field observations and satellite imagery data. The classification was performed in two steps. First, the k-means method was used to determine the number of clusters with minimal intra-cluster variance. Then, the Ward method was applied to identify four clusters representing similar biotopes, namely, fresh clearings, overgrown clearings, pole-stage stands, and young deciduous forest. No clear correlation was found between biotope type and the age of the clear-cut areas. The identified biotopes served as the basis for building a neural network that recognizes these forest types in satellite imagery. To train the neural network model, we used two-thirds of the original data (80 points out of 120 points of descriptions), and one-third for testing. By changing the number of neurons and perceptron layers, we tried to achieve the same error on the training and test samples. In two layers of 25 and 15 neurons each, the required errors equalized amounting to over 80 %. As a result of using the network, a raster grid with five zones and “white denote” was calculated for the entire study area. All computations and cartographic design were carried out in the R programming environment (terra package), while the neural network was developed using the Keras package, which had not been previously used in this area. The resulting map, characterizing environmental factors, will be used in the study of animal populations and their parasites.

Keywords

secondary forests, habitats, remote sensing, neural network, Keras

References

  1. Bespyatova L.A., Bugmyrin S.V., Kutenkov S.A., Nikonorova I.A. The Abundance Ixodid Ticks (Acari: Ixodidae) on Small Mammals in Forest Biotopes of the Middle Taiga Subzone of Karelia. Parazitologiya (Parasitology), 2019. V. 53. No. 6. P. 463–473 (in Russian). DOI: 10.1134/S0031184719060036.
  2. Bugmyrin S.V., Korosov A.V., Ieshko E.P., Anikanova V.S., Bespyatova L.A., Matrosova Y.M., Telegin I.V. The Experience of Studying the Spatial Distribution of Parasites of Small Mammals. Northern Europe in the 21st Century: Nature, Culture, Economy. Proceedings of the International Conference Dedicated to the 60th Anniversary of the Karelian Research Centre of Russian Academy of Sciences (24–27 October 2006). Petrozavodsk, 2006. P. 55–58 (in Russian).
  3. Danilova I.V., Korets M.A., Ryzhkova V.A. Regenerating Vegetation Age Stages Mapping Based on Multi-Seasonal Landsat Satellite Imagery. Earth Research from Space, 2017. No. 4. P. 12–24 (in Russian). DOI: 10.7868/S0205961417040029.
  4. Deur M., Gasparovic M., Balenovic I. Tree Species Classification in Mixed Deciduous Forests Using Very High Spatial Resolution Satellite Imagery and Machine Learning Methods. Remote Sensing, 2020. V. 12. No. 23:3926. P. 1–18. DOI: 10.3390/rs12233926.
  5. Fricker G.A., Ventura J.D., Wolf J.A., North M.P., Frank W.D., Franklin J. A Convolutional Neural Network Classifier Identifies Tree Species in Mixed-Conifer Forest from Hyperspectral Imagery. Remote Sensing, 2019. V. 11. No. 19:2326. P. 1–22. DOI: 10.3390/rs11192326.
  6. Guseva T.L., Korosov A.V., Bespyatova L.A., Anikanova V.S. Long-Term Dynamics of Biotopical Distribution of a Common Shrew (Sorex Araneus, Linnaeus 1758) in Karelian Mosaic Landscape. Scientific Notes of Petrozavodsk State University, 2014. V. 2. No. 8. P. 13–20 (in Russian).
  7. Ieshko E.P., Korosov A.V., Nikonorova I.A., Bugmyrin S.V. Species Richness of Helminth Communities in Relation to Host Abundance Variations (The Case of the Common Shrew Sorex Araneus). Parazitologiya (Parasitology), 2020. V. 54. No. 1. P. 3–12 (in Russian). DOI: 10.31857/S1234567806010010.
  8. Kanev A.I., Tarasov A.V., Shikhov A.N., Podoprigorova N.S., Safonov F.A. Identification of Logged and Windthrow Areas from Sentinel-2 Satellite Images Using the U-net Convolutional Neural Network and Factors Affecting its Accuracy. Current Problems in Remote Sensing of the Earth from Space, 2023. V. 20. No. 3. P. 136–151 (in Russian). DOI: 10.21046/2070-7401-2023-20-3-136-151.
  9. Kedrov A.V., Tarasov A.V. Classification Forest Vegetation with Neural Network. Bulletin of Perm National Research Polytechnic University, 2017. No. 22. P. 44–55 (in Russian).
  10. Korosov A.V. Neural Networks for Ecology: Introduction. Principles of the Ecology, 2023. No. 3. P. 76–96 (in Russian). DOI: 10.15393/j1.art.2023.14002.
  11. Kuzmenko E.I., Frolov A.A., Silaev A.V. Mapping Forest Landscapes of the North-West of Western Siberia Using GIS. Geography and Natural Resources, 2015. No. 4. P. 151–161 (in Russian).
  12. Kurbanov E.A., Vorobyev O.N., Lezhnin S.A., Gubaev A.V., Polevshchikova Yu.A. Thematic Mapping of Vegetation Cover from Satellite Images: Validation and Accuracy Assessment. Yoshkar-Ola: Volga Region State Technological University, 2015. 132 p. (in Russian).
  13. Ma Y., Zhao Y., Im J., Zhao Y., Zhen Z. A Deep-Learning-Based Tree Species Classification for Natural Secondary Forests Using Unmanned Aerial Vehicle Hyperspectral Images and LiDAR. Ecological Indicators, 2024. V. 159. P. 111608. DOI: 10.1016/j.ecolind.2024.111608.
  14. Makhonko Ya.V., Petryaeva A.A., Podmarkova V.A., Galaktionov I.D., Dmitrieva L.A. Review of Methods of Semantic Segmentation of Satellite Images of the Earth Using Neural Network Technologies. System Analysis in Engineering and Control. Proceedings of the XXIX International Scientific, Educational and Practical Conference. SPbPU, 2023. No. 2. P. 258–266 (in Russian). DOI: 10.18720/SPBPU/2/id23-106.
  15. Melnikov A.V., Polishchuk Y.M., Rusanov M.A., Abbazov V.R., Kochergin G.B., Kupriyanov M.A., Baisalyamova O.A., Sokolkov O.I. Comparative Analysis of Neural Network Models for Felling Mapping in Summer Satellite Imagery. Scientific and Technical Journal of Information Technologies, Mechanics and Optics, 2024. V. 24. No. 5. P. 806–813 (in Russian).
  16. Rylskiy I.A. Approaches to the Determination of Taxation Indicators of Forests Using Aerospace Images and Lidar Data. InterCarto. InterGIS. Proceedings of the International conference, 2018. V. 24. P. 216–240 (in Russian). DOI: 10.24057/2414-9179-2018-2-24-216-240.
  17. Scholle F. Deep Learning with R. Moscow: DMK Press, 2022. 646 p (in Russian).
  18. Shitikov V.K., Mastitskiy S.E. Classification, Regression and Other Data Mining Algorithms Using R. Electronic Book, 2017. 351 p. (in Russian). Web resource: https://github.com/ranalytics/data-mining (accessed 01.02.2025).

For citation: Korosov A.V., Marfitsyna N.A. Animal habitat interpretation with KERAS library deep learning methods. InterCarto. InterGIS. Moscow: MSU, Faculty of Geography, 2025. V. 31. Part 2. P. 54–65. DOI: 10.35595/2414-9179-2025-2-31-54-65 (in Russian)