Agriculture Reference
In-Depth Information
0.6
0.5
original spectrum in blue
smoothed spectrum in red
0.4
0.3
1000
1500
2000
2500
wavelength in nm
Fig. 9.10
Soil spectrum as originally recorded and after smoothing or removing noise by discrete
wavelets. The three depressions in the spectral curve are caused by water absorbance (Extracted
from Ge and Thomasson
2006
, altered)
Removing noise can help substantially to get rid of unwanted influences
(Fig.
9.10
) but must be applied with care in order to avoid loss of important informa-
tion. In addition to denoising methods, multiple regression analyses can substan-
tially assist in extracting more information from the spectra such as
•
partial least squares regressions
•
principal component regressions
•
stepwise regressions.
Denoising techniques and regression analyses cannot be dealt with here at length.
In the future, these methods for processing the sensed data will probably often be
used as “
black boxes
” in a similar way as present day computer programs. For
details see Brandt (
2011
), Haykins and Van Veen (
2003
), Martens and Naes (
1992
)
as well as Savitzky and Golay (
1964
). The important point is that a careful applica-
tion of denoising methods and regression analyses makes it possible to extract from
spectroscopic spectra exactly the information that is needed for site-specific preci-
sion farming.
The first step is using the relation between specific soil properties and the reflec-
tance for the
calibration
of the sensor and its software. The reflectance that is
detected in fields of a defined region thus helps to adjust the sensing equipment. The
result can be that for soils that are located outside this defined area, the sensing
device gets inaccurate. However, this is a matter of how much the soil properties
vary geographically. An important question for the future will be, how much gener-
alizing and standardizing of calibration is possible while attempting to use the same
calibration for wide areas.
The calibrated sensing equipment can then be subjected to a
validation
process.
This validation too is a comparison of soil properties as recorded by traditional proce-
dures against the results that were detected by the spectroscopic sensing method. The
difference is that now directly the calibrated instrument is used for the comparison.
