333 
3.1.1. Method. The isolated features are parameterised, the method of parameterisation varying with the 
user, but consisting generally of the absorption position, depth, width at half maximum depth and 
asymmetry. Kruse et al. (1993) explains the process in detail. 
In the case of Kruse et al. (1993) the parameters extracted can then be input into an 
expert system which consists essentially of a set of IF-THEN rules to determine the identity of the 
unknown. The rules base of the expert system needs to be developed by expert analysis of a spectral 
library of minerals that has been parameterised in the same manner. 
3.1.2. Problems of use. Parameterisation can produce accurate mineral distribution maps. However, as 
with all techniques it has associated problems of use which are listed below. 
1. Large number of processing steps 
(a) Feature extraction using hull quotients or equivalent. 
(b) Feature parameterisation. 
(c) Creation of rules base and its modification for additional minerals by expert analysis of a 
2. The sensitivity of the feature extraction step to noise and mixtures in the input data, changing the 
“background” continuum or creating partly overlapping absorption features. 
4. The expert system rules base needs expert input based on a mineral library and may need to be 
modified when additional minerals are added to the library. Also the decision rules are somewhat 
approximate and may not reflect the subtlety of variation detectable between two minerals. 
3.1.3. Advantages of Parameterisation. The major advantages of parameterisation are, 
1. The reduction of the often large reference spectral libraries of minerals with many bands of data to a 
library with few parameters for each mineral. This reduces memory requirements and increases the 
feature, unlike cross-correlation methods which normalise the data and lose this important depth 
information. 
3.2. Cross-correlation 
This technique is widely used in statistics but is a relatively recent introduction as an analytical technique 
for imaging spectrometry (Mackin et al., 1991; Yuhas et al., 1992). Each library mineral spectrum is 
that of Mackin et al. (1991). Prior to the comparison the library mineral spectra are normalised and 
compared to the normalised unknown spectrum. The normalisation removes the multiplicative and additive 
offsets that may confuse identification, such as topographic shading effects in the scene. 
In this method the reflectance spectra are converted to percentage variations (plus or 
minus) about the mean of each spectrum, using, 
parameterised mineral spectral library. 
The identification step itself. 
(d) 
3. Some parameters are sensitive to the noise in the data, for example the absorption maximum depth 
position. 
speed of identification. 
2. The extracted depth parameter can be used to give an indication of the strength of the absorption 
3.2.1. Method. The technique has been implemented in different forms, the description below is based on 
compared against the unknown spectrum across the same wavelength range after converting the library 
spectra to the same bandwidths as the unknown. 
Di = iX L - X s ) 
£i=l ( IXi - XI ) 
£ 1 ) 
Di = Normalised value of band i 
Xi = Reflectance value of band i 
X = Mean value of the spectrum 
n = Number of bands