TWO-LEVEL SYSTEM FOR DECOMPOSITION OF REMOTE SENSING DATA
A. V. Lapko1,2, V. A. Lapko1,2, S. T. Im1,3, Yu. P. Yuronen1
1Reshetnev Siberian State University of Science and Technology, Krasnoyarsk, Russia
2Institute of Computational Modelling SB RAS, Krasnoyarsk, Russia
3Sukachev Institute of Forest SB RAS, Krasnoyarsk, Russia
Keywords: structural analysis of spectral data, automatic classification, correlation coefficient, NDVI method, NDII method, kernel probability density function, statistical data decomposition, remote sensing data, forest area, damaged stands
Abstract
A method for structural analysis of remote sensing data is proposed, based on a two-level decision-making system. At the first level of the system's structure, remote sensing data is decomposed using the signs of the components that make up the estimated correlation coefficient between spectral features. Based on this, four classes of spectral data are identified. At the second level of the system structure, each identified first-level class undergoes further decomposition according to the Euclidean distance values for the components of the correlation coefficient estimate between spectral features. The developed system was used in the structural analysis of remote sensing data of a dark coniferous forest damaged by the Siberian silkmoth. The obtained results were compared with those obtained using the NDVI and NDII spectral indices of remote sensing. Nonparametric probability density estimates were used to analyze the spectral indices of remote sensing data and the developed decomposition system methodology. A procedure for optimizing kernel probability density estimation is considered. Schematic diagrams of the obtained results are provided.
|
