P. Cipollini et al., Retrieval of sea water optically active parameters from hyperspectral databy means of generalized radial basis function neural networks, IEEE GEOSCI, 39(7), 2001, pp. 1508-1524
Citations number
34
Categorie Soggetti
Eletrical & Eletronics Engineeing
Journal title
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING
In this paper, we present a new methodology for estimating the concentratio
n of sea water optically active constituents from remotely sensed hyperspec
tral data, based on generalized radial basis function neural networks (GRBF
-NNs). This family of NNs is particularly suited to approximate relationshi
ps like those between hyperspectral reflectance data and the concentrations
of optically active constituents of the water body, which are highly nonli
near, especially in case II waters. Three main water constituents are taken
into account: phytoplankton, nonchlorophyllous particles, and yellow subst
ance. Each parameter is estimated by means of a specific, multi-input singl
e-output GRBF-NN. We adopt a recently proposed network learning strategy ba
sed on the combined use of the regression tree procedure and forward select
ion. The effectiveness of this approach, which is completely general and ca
n be easily applied to any hyperspectral sensor, is proved using data simul
ated with an ocean color model over the channels of the medium resolution i
maging spectrometer (MERIS), the new generation ESA sensor to be launched i
n 2001. We define the estimation algorithms over waters of cases I, II, and
I+II and compare their performance with that of classical band-ratio, sing
le-band, and multilinear algorithms. Generally, the GRBF-NN algorithms outp
erform the classical ones, except for the multilinear over case I waters. A
particular improvement is over case II waters, where the mean square error
(MSE) can be reduced by one or two orders of magnitude over the error of m
ultilinear and band-ratio algorithms, respectively. We also discuss briefly
, with an example, the noise filtering effects of the network and the effec
ts of the size of the training set.