Agriculture Reference
In-Depth Information
transformations, among others. Some of the most common preprocessing techniques are
presented here. For a full review of preprocessing methods, please refer to Nicolaï et al.
(2007).
2.1.1 Multiplicative scatter correction
Multiplicative scatter correction (MSC) is one of the most commonly used normalization
techniques. In MSC, the light scattering is estimated for each sample relative to an ideal
sample obtained by averaging the complete wavelength range of the data set. Each
spectrum is then corrected such that all samples appear to have the same scatter level as the
reference spectrum. For a further description of this method, see (Isaksson & Naes 1988;
Naes et al. 1990; Vohland et al. 2009).
2.1.2 Absorbance
Spectroscopic measurements performed in transmission mode can be quantified using
Beer's Law (concentration
∼
log(1/
T
) where
T
is the transmitted radiation). Accordingly,
reflectance measurements are frequently converted to log (1/
R
) values, which are then used
in a manner similar to optical density readings (Nicolaï et al. 2007).
2.1.3 Kubelka-Munck
The Kubelka-Munck transformation, (1-
R
2
)/2
R
,
is similar to absorbance but accounts for
scattering (Nicolaï et al. 2007).
2.1.4 Continuum removal
The continuum is the background absorption on which other absorption features are
superimposed. The spectrum is divided by a convex hull that is fitted over it (Clark & Roush
1984; Clark & others 1999).
2.1.5 Derivatives
Derivation is often used to remove baseline shifts and superposed peaks. Second-derivative
spectra can correct for both additive and multiplicative effects (like MSC). They are usually
calculated according to the Savitzky-Golay algorithm (Naes et al. 2002). The parameters of
the algorithm (interval width, polynomial order) should be carefully selected to avoid
amplification of spectral noise (Nicolaï et al. 2007).
2.1.6 Spectral mixture analysis
Spectral mixture analysis (SMA) is a widely used method to determine the sub-pixel
materials that fundamentally contribute to the spectral signal of mixed pixels. This is of
particular importance for obtaining quantitative estimates of distinct materials, a typical
application of remote sensing hyperspectral data. SMA aims to decompose the measured
reflectance spectrum of each pixel into the proportional spectral contribution of so-called
endmembers (EMs). In recent years, many authors have proposed and used a more complex
model, in which both the number and the set of EMs vary dynamically on a per-pixel basis;
this has become known as multiple EM SMA (MESMA). The idea consists of restricting the
large set of possible EMs to a small set of better suited EMs, which can be different for each
pixel, thereby allowing an accurate decomposition using a virtually unlimited number of
