2-2 REVERSE MATCHING FROM MAPS TO ORIGINAL IMAGE 
Because the resampling of TM image in fine processing lead to 
degrading of the spectral intensity, it is not beneficial to the 
computer classification and visual interpretation. In order to 
restore spectral information in original image, we make reverse 
matching from maps to the rough processing image.For this purpose 
reverse matching should be done with 
NL=F’(L,P) NP=F’(L,P) (2) 
in which L,P and NP,NL are coorinates of fine and rough processing. 
The digital maps are also transformed into rough processsing 
image during reverse marching. Diagram of reverse matching see 
FIGURE2. 
2- 3 OUTPUT PRODUCTS AND THEIR ACCURACY 
Two kinds of color TM image format maps are produced from com 
puter system. they are MATCH image format map. And REVERSE MATCH 
image format map. 
The residuals of match image format map after 3rd— order 
polynomial transformation are <5x= 25.93m, 5y= 24.67m (see table 1) 
The standard error on topographic maps are <5x=5y= 25m.The total 
RMSRs are <5x= 36.02m,Sy=35.14m. after randomly selecting 14 check 
points the standard error are 5x= 31.17m,5y= 17.11m. 
Tb accuracy of reverse match image format map is about 1.5 
pixel, which can meet requirment of computer classification and 
evaluation. 
3 APPLICATION OF MULTI-DATA IN CLASSIFICATION OF GRASS RESOURCES 
3- 1 GRASSLAND TAXONOMY IN THE SOUTH OF CHINA 
The traditional taxonomy of the grassland in south part of china 
can be classed at 3 levels — CLASS, GROUP, TYPE. 
At the .first level grassland is divided into 5 CLASSes: 
(1) Grassland (2) Bush-grassland (3) Wood-grassland 
(4) Meadow (5) Odd pieces of glassland 
there are only (1), (2), (3) class in Lichuan county. 
At the second level every class is divided into 3 GROUPs: 
(1) High-mountain group,in which the terrain is higher than 1200m. 
(2) Mid-mountain group, in which the terrain is between 800m and 
1200m. 
(3) Low-mountain group, in which the terrain is lower than 800m. 
At the third level grassland-TYPEs are determined on the basis 
of grass format. 
3-2 CLASSIFICATION 
3-2-1 Classification based on spectral feature 
The spectral bands TM3,TM4 and TM5 are used in classification 
of spectral feature. They are determined in feature selection. The 
algorithm of classification is maximum likelyhood method: 
Di (x)=P(x/i)P(i) (3) 
P(x/i)= - EXP(-i (X-Mi) T £“} (X-Mi)) (4) 
|Ei | 1/2 (2tt) K/2 2 
195