International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B7. Istanbul 2004 
  
Feature Oriented Principal Components Selection is based on the 
examination of PCA eigenvector loadings to decide which of the 
principal component will extract information directly related to 
the theoretical spectral signatures of specific targets. The 
methodology relies specifically on the selective input of only four 
image bands for PCA. This technique performed by using four 
selected TM bands in order to highlight the spectral response of 
iron-oxide minerals (absorption in visible TM bands 1 & 2 and 
higher reflection in TM3) and hydroxyl-bearing (clay) minerals 
(absorption in TM7, higher reflectance in TM5). 
  
  
Yenipinar 
  
  
  
  
  
  
  
Figure 4. Color composite of Crosta H, H+F and F as RGB. 
2.1.4 Least Squares Fitting Method: The technique assumes 
that the bands used as input values are behaving as the variables of 
a linear expression. And the ‘y’ value of the equation, namely the 
predicted band information, gives us a calculated output value. 
This predicted band is what that band should be according to the 
linear equation. The problem of having vegetation responsible of 
some reflectance in the bands that are used to map clay minerals 
can therefore be omitted by using this technique. The vegetation is 
mapped in the predicted band with the values that are calculated 
just by using the reflectance information in the other bands. The 
minerals which are sensitive to a specific band are then 
differentiated from the features which are reflective to the other 
bands as well; just by taking the difference between the predicted 
values and the original values. Calling this difference, the residual, 
color composites are displayed with the specific anomalously 
reflective features and then interpreted (Fig. 5) (Clark et al., 
1990). 
2.2 Mineral Mapping by using Spectral Reflectance Data 
The technique is purely based on the band rationing process. More 
than the band selection criteria, a filtering script based on the 
statistical calculations is developed. The technique can be 
summarized as in a flow chart in Figure 6. 
Spectral reflectance data of the alteration minerals is statistically 
processed according to the TM band intervals. Descriptive 
statistics like the minimum, maximum, mean and standard 
deviation are calculated for the TM band intervals separately for 
every mineral. 
388 
  
  
Utuóren 
  
  
  
  
  
Figure 5 Color composite of LS-Fit. residual bands 3, 7 and | as RGB. 
P-P and Q-Q plots are analyzed to obtain information about the 
distribution type of the data belonging to the band intervals for 
every mineral. Generally data is found to be normally distributed. 
For all of the data it is therefore assumed to be normally 
distributed, although it is not an accurate assumption for the whole 
data. 
Two standard deviations are added to the mean value to have the 
upper limit and subtracted to have the lower limit of the mineral 
mask that is used to filter the band ratios. 
The analysis performed in this section includes the processing and 
filtering of selected alteration minerals which are Kaolinite, Illite, 
Montmorillonite, Pyrophyllite, Alunite, Orthoclase, Quartz, 
Epidote, Chlorite, Hematite, Goethite and Jarosite. Basically band 
ratio technique is applied and filtered according to the statistically 
calculated ratio intervals for. these minerals. Band ratios are 
selected according to the spectral curves, band combinations that 
will give distinctive ratios (very high or very low) are preferred. In 
Figure 7 Kaolinite pixels are mapped that have passed from the 
filter. 
2.3 Accuracy Assessment 
The pixels in the reference image and the image obtained after 
spectral analysis method are correlated to calculate the percentage 
accuracies (Table 3). Counted pixel values are shown in the table. 
Accordingly the (1,1) value is the pixels both mapped in the 
original data and in the spectral analysis result; (0,1) is the pixels 
displayed by original and missed by the analysis; (1,0) pixels are 
displayed only by analysis result and not exist in the original data 
and value of (0,0) are the pixels both did not displayed by original 
data and analysis result. Overall accuracy is calculated as % 70.7, 
Correlation Matrix 
Total Commission 
15814 76 
Result 65535 17 
57,32 8,15 
Omission 
42.68 91,85 
Table 3. Accuracy assessment matrix 
    
  
  
Internai 
r^ 
  
  
Figure 7. 
Kaolinite 
All of the 
centers of 
complexes 
minerals an 
PCA and L: 
The propo: 
individually 
and found t 
mineral 30 
likely to gi 
Although s 
not be map 
previously ; 
as outputs o