Integration of spectral channels in the classification of coniferous and deciduous vegetation from satellite images

Abstract

The results of the study of the classification procedure of coniferous and deciduous, as well as coniferous and mixed (in equal percentages) vegetation according to Landsat–7 images for different seasons of the year are presented. Spectral channels with a spatial resolution of 30 meters were used. Vegetation classification was carried out by brightness characteristics using the nearest neighbor method. The reference brightness of objects (simple standards) was determined by their mathematical expectations in each spectral channel for each season. Additionally, aggregated standards are formed by combining the brightness of plant objects in spectral channels. It is shown that when separating coniferous and deciduous objects, the probability of correct selection of coniferous can reach 1.0000 when using simple standards. At the same time, the probability of correct allocation of deciduous does not exceed 0.9697. The use of aggregated standards makes it possible to increase this probability to 0.9899. When classifying coniferous and mixed vegetation, the effectiveness of aggregated standards turned out to be lower than that of simple ones selected by spectral channels and shooting seasons. The obtained results suggest the continuation of research when dividing plant objects into a larger number of classes.