Войти на сайт
МЕЖДУНАРОДНЫЕ ЕЖЕГОДНЫЕ КОНФЕРЕНЦИИ
"СОВРЕМЕННЫЕ ПРОБЛЕМЫ ДИСТАНЦИОННОГО
ЗОНДИРОВАНИЯ ЗЕМЛИ ИЗ КОСМОСА"
(Физические основы, методы и технологии мониторинга окружающей среды, природных и антропогенных объектов)

Восемнадцатая Всероссийская Открытая конференция «СОВРЕМЕННЫЕ ПРОБЛЕМЫ ДИСТАНЦИОННОГО ЗОНДИРОВАНИЯ ЗЕМЛИ ИЗ КОСМОСА (Физические основы, методы и технологии мониторинга окружающей среды, потенциально опасных явлений и объектов)»

XVIII.A.504

Deciduous tree species identification using airborne hyperspectral data and phenological stages

Brovkina O. (1,2), Pujiastuti I. (3), Fabiánek T. (1), Lagutov V. (3)
(1) Global Change Research Institute CAS, Brno, Czech Republic
(2) St.Petersburg Institute for Informatics and Automation RAS, Saint Petersburg, Russia
(3) Central European University, Budapest, Hungary
The study explore an optimal phenological period to identify 9 tree species in mixed deciduous floodplain forest in the Czech Republic: Austrian oak (Quercus cerris L), wild apple (Malus L.), conker tree (Aesculus hippocastanum L.), hornbeam (Carpinus betulus L.), ash (Fraxinus angustifolia L.), maple (Acer canpestre L.), elm (Ulmus laevis L.), linden (Tilia europaea L.), willow (Salix carpea L.). Airborne hyperspectral data (spectral range of 380-1050 nm, spectral resolution of 9 nm, spatial resolution of 1 m) were acquired six times during a vegetation period: end of April, beginning of July, end of July, beginning of August, beginning of September and mid of October. Maximum Likelihood, Artificial neural network (ANN) and Support vector machine (SVM) methods were applied to classify tree species. The spectra analysis showed that the most important spectral variables to distinguish between 9 tree species were located in red (520..580 nm) and near infrared (740..1050 nm) spectral regions. Individual tree species accuracies ranged from 28.06% (wild apple) to 93.1% (conker tree). Maximum Likelihood achieved higher Kappa coefficient (K) in each phenological period than other methods, with that the Overall accuracy (OA) and Average accuracy (AA) were not high using Maximum Likelihood. SVM achieved a higher level of classification accuracy (OA = 89.4%, AA = 81%, K = 0.78) than other methods in the beginning of July, that was found as the optimal phenological period to identify 9 tree species in mixed deciduous forest.
The study was supported by the Ministry of Agriculture of the Czech Republic grant number QK1910150.

Ключевые слова: classification, image spectroscopy, deciduous forest, SVM, ANN, maximum likelihood
Литература:
  1. Grigorieva O., Brovkina O., Saidov A. 2020. An original method for tree species classification using multitemporal multispectral and hyperspectral satellite data. Silva Fennica 54 (2).
  2. Hanuš, J., Fabiánek, T., Fajmon, L. 2016. Potential of airborne image spectroscopy at CzechGlobe. The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XLI-B1.
  3. Richards, J.A. Remote Sensing Digital Image Analysis Berlin: Springer-Verlag (1999), 240 pp

Методы и алгоритмы обработки спутниковых данных

4