changes little, the de-noising effect is good, and the original radiance data image was converted to reflectance data image.
The whole scene preprocessed image was used as the experimental data (hereinafter referred to as the BD image), as
shown in Figure 1 (a).
(a)BD Image (b)AD Image
Figure1. Experimental Image(R:786.1nm,G:593.2nm,B:511.5nm, false color composite)
2.2 Field work
The image coverage area is located at the junction of the old and new river channel in the Yellow River Estuary. The
object types in this area are rich, natural wetland including reed, seepweed, tamarisk and tidal flat, artificial wetland
including aquaculture water, reservoir and pond. In order to evaluate the experimental classification accuracy accurately,
the author's research group carried out the field reconnaissance in September 2012, stroked 48km, using the method of
typical sample area and route records to collect hyperspectral data of object types, more than 100 groups of hyperspectral
data and 230 photos were acquired. Using the interactive interpretation guided by the field work data, the interpreting
images of the CHRIS image covering area were generated (Figure 4), in which red for reed wetland, yellow for bare
beach, blue for clear water (seawater and pond), green for turbid water (the Yellow River), cyan for seepweed, purple for
tamarisk.
2.3 Derivative transformation
Derivative transform is a nonlinear mathematical transformation method. It can effectively highlight the slope
characteristics of the spectral curves, allowing researchers get easier to compare the spectral characteristics of the object
spectral curve. In the mages after derivative transformation, the characteristic information of spectrum is highlighted, and
is beneficial to the land cover identification and classification. In this study, the first-order derivative algorism is
estimated by Eq. (1)
|
=
[5]
(1)
Where
λ
i
and
λ
j
is the center wavelength of image bands, s(
λ
)is the true signal spectrum of the band
λ
, ∆λ is the
separation between adjacent bands,
∆λ=
λ
i
-
λ
j
, and
λ
i
>
λ
j
.
The experimental image data was transformed using the first-order derivative transformation formula of Eq. (1), the
transformed image (hereinafter referred to as the AD image) as shown in Figure 1 (B). After derivative transform, the
Proc. of SPIE Vol. 9142 91421O-3