A NEW PERSPECTIVE TO VISIBLE NEAR INFRARED REFLECTANCE

1
A NEW PERSPECTIVE TO VISIBLE NEAR INFRARED
REFLECTANCE SPECTROSCOPY A WAVELET APPROACH
Yufeng Ge, Cristine L.S. Morgan, J. Alex
Thomasson and Travis Waiser Texas AM University,
College Station, TX
INTRODUCTION
MATERIALS AND METHODS CONT
Partial Least Squares (PLS) Analysis
Justification

• Spectra were averaged every 10 nm Waiser et al.,
  200X
• The 1st derivative was used for model building
  Waiser et al., 200X
• Partial least squares regression was performed
  using Unscrambler 9.0 (CAMO Tech, Woodbridge, NJ)

In soil science, visible and near-infrared
diffuse reflectance spectroscopy is being used in
an effort to develop proximal sensors to quantify
soil constituents. Common analysis techniques in
soil spectroscopy include principal component
analysis, partial least squares regression (PLS),
and boosted regression trees. These techniques
limited ones ability to assess wavebands
important to prediction models. A new algorithm
to incorporate wavelet analysis into VNIR
spectroscopy as a preprocessing tool is proposed
in this study. The technique uses a discrete
wavelet transform (DWT) to analyze soil
reflectance at multiple spectral resolutions.
RESULTS
Objectives

• Compare the results of using DWT regression to
  the results of PLS regression in predicting clay
  content using VNIR scans of in-situ soil cores.
• Examine the effectiveness using the results of
  the two models for physical interpretation of
  results.

MATERIALS AND METHODS
Soil Scanning and Clay Analysis

• 72 soil cores were collected from 21 soil series
  in Central Texas. Soil parent material varied
  from residuum to alluvial materials

Wavelet w/ Multiple Regr.
waveband scale
mica smectite kaolonite

• Soil cores were cut open and scanned using an ASD
  FieldSpec Pro FR VNIR spectroradiometer
  (Analytical Spectral Devices, Boulder, CO), with
  a spectral range of 350-2500 nm

iron oxides
351
2398
1374
862
1886

• The pipette method was used to measure particle
  size distribution of the soil samples

waveband (nm)
CONCLUSIONS

• 70 of the cores were used for model calibration,
  30 of the cores were used for model validation

• DWT and PLS regression predicted clay content
  with similar accuracies, the PLS prediction was
  slightly better
• The DWT model creates a simple diagram for
  visualizing central wavelengths and scales of
  wavelet regressors, facilitating physical
  interpretation of the prediction model.
• Spectral maps of significant regressors in both
  methods indicate that iron oxides, clay minerals
  and soil color are used in the clay content
  prediction models
• The wavelet method is more suitable for
  inexpensive sensor development because it would
  allow for a simpler sensor design.

Discrete Wavelet Transform (DWT) Analysis

• A dyadic discrete wavelet transform, the Haar
  wavelet, was used for its simplicity Ge et al.,
  200X
• Spectra were truncated to 2048 data points with
  the spectral range from 351 2398 nm
• Each soil spectra was subject to seven levels of
  dyadic filter band decomposition at scale 3, 4,
  5, and 6 (or spectral band ranges of 256, 128,
  64, and 32 nm, respectively)
• Stepwise multiple linear regression was used with
  the wavelet variables to develop prediction
  models for soil clay. The p-value was set at 0.05
  for a regressor to be added, and at 0.1 for a
  regressor to be removed
• The DWT and stepwise multiple linear regression
  was performed with Wavelet Toolbox and Statistics
  Toolbox, respectively, in MATLAB Release 13

REFERENCES

• Ge, Y., C.L.S. Morgan, J.A. Thomasson, and T.H.
  Waiser. 2006. A new perspective to near infrared
  reflectance spectroscopy A wavelet approach.
  Trans. ASAE. Submitted.
• Waiser, T., C.L.S. Morgan, D.J. Brown and C.T.
  Hallmark. In situ characterization of soil clay
  content with visible near-infrared diffuse
  reflectance spectroscopy. SSSAJ. Submitted.