RECOGNITION OF AGRICULTURAL LANDS BASED ON MEASUREMENTS OF A VEGETATION INDEX

  • Sergey Aleksandrovich Bartalev Space Research Institute for the Earth
  • Alexander Leonidovich Zakora Space Research Institute for the Earth
Keywords: VEGETATION INDEX NDVI, CLASSIFICATION OF AGRICULTURAL LAND

Abstract

The paper presents a method for assessing the utilization of arable land based on measurements of the NDVI vegetation index, calculated from images of a MODIS spectroradiometer from the Terra satellite. The work involves the development of a software module that performs the automatic classification of agricultural land into utilization classes.An algorithm and programmatic block are worked out for classification of degree of used of agricultural lands. Educating of classifier is conducted by means of algorithms of computer-aided instruction.

For an analysis data of measuring of vegetation index of NDVI were used for every week 2017 year, got in the system Vega the vehicle of MODIS. For testing of algorithms a selection was used from 1026 vectors of signs (agricultural lands) equipartition on different areas. Breaking up on teaching and test selections was produced with maintenance of balance between areas. A job of the program performance is a mark of class for every entrance vector, characterizing the use of agricultural lands. The use of measuring of vegetation index allowed to automatize classification of used of the fields. For the fields, vector of values of NDVI that is had admissions of measuring, the algorithm of filling of admissions is applied, that allowed to use these the vector for classification. The features of development of vegetation were taken into account depending on a geographical location. The use of algorithms of computer-aided instruction gave next results: kNN is exactness of classification of 82%; SVC is exactness of classification of 78%; Random Forest is exactness of classification of 85%; GBT is exactness of classification of 86.307%. The most high results are got with the use of algorithm of GBT - 86% of the correctly classified fields. Programmatic block realized as a module ready for integration in the system Vega.

References

Барталев С.А., Егоров В.А., Жарко В.О., Лупян Е.А., Плотников Д.Е., Хвостиков С.А., Шабанов Н.В. Спутниковое картографирование растительного покрова России: М. ИКИ РАН, 2016. 208 с.

Барталев С.А., Егоров В.А., Лупян Е.А., Плотников Д.Е., Уваров И.А. Распознавание пахотных земель на основе многолетних спутниковых данных спектрорадиометра MODIS и локально-адаптивной классификации. Компьютерная оптика, 2011. Т. 35. № 1. С. 103–116.

Стыценко Е.А. Возможности распознавания сельскохозяйственных угодий с использованием методики совместной автоматизированной обработки разносезонных многозональных космических изображений. Современные проблемы дистанционного зондирования Земли из космоса, 2017. Т. 14. № 5. С. 172–183

Лутц, М. Программирование на Python, II том: М. Лутц. - СПб: Символ-плюс, 2015.

Max Kuhn, Kjell Johnson. Applied Predictive Modeling: Springer, 2019.

ВЕГА-Science [Электронный ресурс]: URL: http://sci-vega.ru.

Закора А.Л., Барталев С.А. Распознавание пахотных земель на основе измерения индекса NDVI //Материалы XXI Международной молодежной научно-практической конференции «Человек и Космос» (10 -12 апр. 2019г.) - Днепр: Национальный центр аэрокосмического образования молодежи им. Макарова, 2019. - DVD, ISSN 2221-4550.

Published
2019-12-30
How to Cite
Bartalev, S. A., & Zakora, A. L. (2019). RECOGNITION OF AGRICULTURAL LANDS BASED ON MEASUREMENTS OF A VEGETATION INDEX. Journal of Rocket-Space Technology, 27(4), 3-8. https://doi.org/10.15421/451901