Graphical Analysis of Spectral Reflectance Curve
This curve can be than described by some graphical invariant. It is possible to establish three invariant group.
Invariant group I: this group consists of only one invariant. This is shape or modulation of the spectral reflectance curve. This invariant is defined by 15 parameters that simulate modulation of spectral curve. These 15 parameters are defined as follows.
M
ij Relation between band j and band I, I= 2,3,4,5,6 and j= 1,2,3,4,5
The value M
ij is assigned 1 when bi<b
j, 2 when b
i=
b
j and 3 when b
i>b
j. from theoretical point of view, there would be 3 power 15 possible combinations of M
ij. If we assign each combination one integral value then this value will range form 0 to 14348907. in practice the number of combination is far less this value. Generally, for n channel data set we need 9n(n-1)/2 parameter to define the modulation of spectral curve. If this invariant will be used for image classification then the number of categories will be 3 power (n(n-1)/2 in maximum number of spectral reflectance curve patterns depends very much on geometric correction processing. The best result will be obtained with nearest neighbor resampling while other method based on interpolation will falsely increase number of spectral reflectance curve patterns
Invariant group II: there group is composed of several invariant that are defined as angles among different segments of spectral reflectance curve. For example angle a on figure 2 can be considered as one of the invariant.
Figure 2: Definition of invariant
Invariant group III: there are several values that may be used as invariant of this group. They are : bands ratio , indices ….. The author has tried with the following values:
- Normalised vegetation index to classify vegetation coverage.
- Area of the polygon limited by the spectral reflectance curve and x-axis ( doted area on the figure 2).
Classification example
To demonstrate the proposed algorithm, the author used a window of LANDSAR TM scene 127/45 observed on 16 October 1996 with 6 channels. The channel 6 stands of the original TM channel 7. the size of the window is 1024 pixels x 1024 lines . on figure 3 is false colour composite of the study are RGB= 432. the study area cover the Hanoi city and vicinity. Land cover of this are is featured by major categories as built up area, water body, paddy field, bare soil and their mixture. The target of the research is to carry out implementation of the proposed algorithm. The author has wrote program composed of Microsoft Visual C 4.0 and FORTRAIN Power station 4.0 modules under windows 95 operating system environment for classification of the extracted window. The computation in divided into the following steps:
Figure 3: False colour composite of the study area
-
Image encoding to get look up table of all unique pixel vectors .
- Classification of LUT to find out all spectral reflectance curves patterns within the images .
- Classification of spectral reflectance curve by GASC algorithm
- If number of patterns is more than 256-> reduction by using frequency value to achieve one bytes classified image
If we will take into account only spectral reflectance patterns with frequency higher than 5 then number of patterns of this window is 171. due to code system one is relative and the other is absolute. The relative code of each spectral reflectance patterns ranges from 0 to 255 and changes from image while the absolute code ranges from 0 to 14348907 and is unique for each land cover category.
