the 
ch 
me 
ial 
for 
by 
1e 
1e 
1€ 
nt 
of 
nt 
at 
Bethel, James 
  
U 
Xu =(Z-Z)—+X, (12) 
. W, 
y (Z-Z y TY 
enm Mehr as Ty (13) 
W, 
where [U, j^ Wl = M" |x y - Al 
From equation (12), Z is a known coordinate of the check point. The residuals are then the differences between the 
estimated and known values. 
X Ay X (14) 
Ya Aa (15) 
where X,Y are the known check point ground coordinates 
2.4.1 Contribution of Control Lines. Experiments were run to test the contribution of linear features with the first and 
second order Gauss-Markov model for varying control features. First, only control points were used to determine the 
unknown parameters. Next, the experiments were repeated with control lines in addition to control points. For the first 
order Gauss-Markov model, two data sets showed similar results. Adding straight line features to the solution improves 
AY (cross track) of check points and check lines significantly, while AX (along track) of check points were essentially 
same: see Tables 1. The second order Gauss-Markov model showed similar results except AX of the check points for 
the Washington, DC data set (Tables 1). Adding control lines reduced the RMS residual AX of check points of the 
Washington, DC data set significantly. The results from the first order GM model slightly better than those from the 
second order GM model. 
Table 1 Contribution of Control Lines 
  
  
  
  
  
  
  
  
  
Check Point RMS (m); Check Point RMS (m); 
GM Control Data Washington, DC Fort Hood 
AX AY Ad AX AY Ad 
Point only 1.95 1.99 2.79 1.61 2.30 2.81 
First order | Point and Line 1.98 1.61 2.55 1.50 1.32 2.00 
Second Point only 2.50 2.14 3.29 1.67 1.53 2.26 
order Point and Line 2.01 1.59 2.56 1.61 1.25 2.04 
  
  
  
  
  
  
  
2.4.2 Comparison of Line Models. Two different line models were tested and compared in table 2. The results of the 
control line model is significantly better than those of the constrained line model for the RMS residual AY with both of 
data sets, while the difference of AX of check points was insignificant. 
Table 2 Comparison of Line Models 
  
  
  
  
Check Point RMS (m); Check Point RMS (m); 
Line Model Washington, DC Fort Hood 
AX AY Ad AX AY Ad 
Control Line 1.98 1.61 2.55 1.50 1:32 2.00 
Constrained Line 1.93 2.02 2.79 1.64 1.74 2.39 
  
  
  
  
  
  
  
  
  
3 EXTRACTION OF THEMATIC CLASSES BY MULTISPECTRAL ANALYSIS 
Hyperspectral image data such as that produced by the HYDICE sensor system provide a rich source of spectral 
information in image form that can be easily exploited in the task of generating a thematic map of an area. The more 
than 200 samples of the spectrum of reflected energy provide ample detail of the spectral response reflected from a 
  
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B7. Amsterdam 2000. 187