The article substantiates the relevance of the study of automated methods for determining the taxational signs of forest plantations based on remote sensing data in the interests of Rosneft. In this work the study of the possibilities of using materials obtained from unmanned platforms to automate the determination by species composition of woody vegetation was continued. As the main approach, an object-oriented approach was applied to classification using extracted statistics on spectral features. The practical implementation of the determination by species composition described in the article was carried out in several stages. Preliminary processing of aerial survey materials with the DJI P4 Multispectral and DJI Matrice 600 Pro UAVs was performed. Training sample was formed based on the reference elements in the test polygon. The segmentation was performed by buffering the vertices of trees obtained at the first stage of processing by laser reflection points. Statistics are collected on the channels of all input images (a data set of 4 rasters with a different combination of channels). The models were trained and classified using the controlled Support Vector Machines method across the entire set of created multi-channel images. Comprehensive assessment of the accuracy of the Support Vector Machines algorithm on the input data was carried out, subsequently, the Random Forrest algorithm was used for comparative analysis on the most informative image with subsequent evaluation of the results. Conclusions are made about the conditions of applicability of both methods, increasing the information content of multispectral images by using signal reflection intensity as an additional data channel, achieving an overall accuracy level of 84% with the combination of data and the Support Vector Machines algorithm recommended in the article, and promising areas for further research are also identified.
References
1. De Luca G., Silva M.N., Cerasoli S. et al., Object-based land cover classification of cork oak qoodlands using UAV Imagery and Orfeo ToolBox, Remote Sensing, 2019, no. 11(10), DOI: 10.3390/rs11101238.
2. Blokhinov Yu.B., Gorbachev V.A., Rakutin Yu.O., Nikitin A.D., A real-time semantic segmentation algorithm for aerial imagery (In Russ.), Komp'yuternaya optika, 2018, V. 42, no. 1, pp. 141–148.
3. Abhishek A., Minakshi K., Raghavendra S., An integrated object and machine learning approach for tree canopy extraction from UAV datasets, Journal of the Indian Society of Remote Sensing, 2021, January, DOI: 10.1007/s12524-020-01240-2
4. Onishi M., Takeshi I., Explainable identification and mapping of trees using UAV RGB image and deep learning, Scientific Reports, 2021, no. 11(1), DOI: https://doi.org/10.1038/s41598-020-79653-9
5 Pogorodniy A.N., Filin N.N., Shumeyko S.A. et al., The unmanned aerial vehicles usage experience on tasks of forest inventory and topography (In Russ.), Neftyanoe khozyaystvo = Oil Industry, 2021, no. 9, pp. 90–94, DOI: https://doi.org/10.24887/0028-2448-2021-9-90-94
6. Filin N.N., Pogorodniy A.N., Arbuzov S.A., Berdnikov N.N., Using aerial survey materials in order to determine the rock and altitude components of the characteristics of forest plantations when conducting surveys at the facilities of Rosneft (In Russ.), Neftyanoe khozyaystvo = Oil Industry, 2022, no. 9, pp. 111–115, DOI: https://doi.org/10.24887/0028-2448-2022-9-111-115
7. Cho M.A., Skidmore A.K., A new technique for extracting the red edge position from hyperspectral data: the linear extrapolation method, Remote Sensing of Environment, 2006, no. 101, pp. 181–193, DOI:10.1016/j.rse.2005.12.011
8. Tolkach I.V., Saevich F.K., Spectral and brightness characteristics of the main forest-forming species on images of the scanner Leica ADS100 (In Russ.), Trudy BGTU, 2016, no. 1, pp. 24–27.
9. Kurbanov E.A., Vorob'ev O.N., Gubaev A.V. et al., Assessment of accuracy and comparability of forest cover thematic maps of different spatial resolution by example of Middle Povolzhje (In Russ.), Sovremennye problemy distantsionnogo zondirovaniya Zemli iz kosmosa, 2016, V. 13, no. 1, pp. 36–48, DOI: 10.21046/2070-7401-2016-13-1 36-48
10. Yunusov A.G., Dzhdid A.D., Beglyarov N.S., Elshevi M.A., Assessment of automatic segmentation accuracy with various point cloud density (In Russ.), Geodesy and cartography = Geodezia i Kartografia, 2020, V. 81, no. 7, pp. 47–55, DOI: 10.22389/0016-7126-2020-961-7-0-0
11. Di Martino M., Hernandez G., Fiori M., Fernandez A., A new framework for optimal classifier design, Pattern Recognition, 2013, V. 46. no. 8, pp. 2249–2255, DOI:10.1016/j.patcog.2013.01.006
12. Ling M., Cheng Q., Peng J. et al., Image semantic segmentation method based on deep learning in UAV aerial remote sensing image, Hindawi Mathematical Problems in Engineering, 2022, pp. 1–10, DOI: 10.1155/2022/5983045
13. Osco L.P., Marcato J. Jr., Ramos A.P.M. et al., A review on deep learning in UAV remote sensing, International Journal of Applied Earth Observations and Geoinformation, 2021, V. 102, DOI: 10.1016/j.jag.2021.102456.
