Sovremennye problemy distantsionnogo zondirovaniya Zemli iz kosmosa, 2010, Vol. 7, No. 4, pp. 112-117
Using the neural network algorithm for determination
of qualitative features of remote sensed Earth surface
on the base of multispectral and radar data
A.A. Ilyin
1, V.V. Egorov
2, A.P. Kalinin
3, I.D. Rodionov
41 oint Stock Company, Research and Development Center Reagent, 119991Moscow,4 Kosygina str
2 Space Research Institute RAS, 117997 Moscow, 84/32 Profsoyuznaya str
3 A.Ishlinsky Institute for Problems in Mechanics RAS, Russia 119526 Moscow, bld. 1, 101 Vernadskogo avenue
4 N.N.Semenov Institute of Chemical Physics, 119991Moscow,4 Kosygina str
It is proposed to use neural networks for definition of qualitative features of remote sensed Earth surface on the base
of multispectral and radar data. The received in situ data were divided on the train and test one. These data were
used for the neural networks training and following inverse problem solution. The verification of neural network
algorithms demonstrated rather high quality - clustering analysis accuracy was equal 0.98 and root-mean-square
error of the cannabis biometric parameters determination was in the range of 3-6%.
Keywords: multispectral survey, remote sensing, neural network, training, verification, accuracy, in situ data
Full text