Since introduction of neural networks into remote sensing they demonstrate good efficiency in remote sensing data analysis.This work is devoted to processing of multispectral(12 bands)images from Sentinel-2(A,B)satellites.Satellite images of areas in Krasnoyarsk Region and Khakassia with known vegetation types are used as task books to train neural networks.Trained neural networks have been reduced to determine which bands are significant for vegetation type identification.Reduction of trained neural network show that vegetation type can be determined from only four infrared bands without significant loses in performance in comparison with non-reduced neural network.
展开▼
