Semi-automatic land cover classification using ADEOS/AVNIR multispectral data
GASC algorithm for four spectral data set
The four spectral bands of AVNIR sensor are equipment to blue, green, red and infrared waves. Given a pixel vector
P(b
1,b
2,b
3,b
4)
b
1 Normalized digital count of band Mu 1
b
2 Normalized digital count of band Mu 2
b
3 Normalized digital count of band Mu 3
b
4 Normalized digital count of band Mu 4
then its spectral reflectance curve can be drafted as follows
Fig. 1: Definition of spectral reflectance curve.
For ADEOS/AVNIR data band 1, band 2, band 3 and band 4 refer to blue, green, red and infrared band. In order to describe shape of spectral reflectance curve defined by four spectral bands we need to develop several invariant.
Delta invariant: This invariant is defined by six parameters that simulate modulation of spectral curve. These six parameters are as follows:
M
1 Relation between band 2 and band 1.
M
2 Relation between band 3 and band 1.
M
3 Relation between band 3 and band 2.
M
4 Relation between band 4 and band 1.
M
5 Relation between band 4 and band 2.
M
6 Relation between band 4 and band3.
The value M1 is assigned 1 when
b
2<b
1, 2 when b
2=b
1 and 3 when b
2>b
1.
The value of M
i for the other band pairs is defined by similar way.
From theoretical point of view, there would be 3 power 6 possible combinations of M
i. If we assign each combination one integral value then this value will range from 0 to 728. In practice the number of combination is far less then this value and is about 74 only.
Alpha invariant: This is angle
a defined as on the Fig. 2
Beta invariant: This is angle
b defined as on the Fig. 2
Gamma invariant: This invariant is defined as an area of the polygon ABCDEF on Fig.2
Figure 2 Definition of Invariant alpha, beta and gamma.
