Identification of western Canadian wheat classes at different moisture levels using nearinfrared NIR

1
Identification of western Canadian wheat classes
at different moisture levels using near-infrared
(NIR) hyperspectral imaging

• S. Mahesh, D.S. Jayas, J. Paliwal, and N.D.G.
  White
• CSBE Annual Meeting 2008

2
Outline

• Introduction
• Objectives
• Materials and Methods
• Results and Discussion
• Conclusions and Future work
• Acknowledgements

3
Introduction

• Wheat production 26.7 Mt and
• export 14.0 Mt in Canada in 2005
• (FAO statistics)
• Eight major wheat classes in western Canada
• Canada western red spring (CWRS)
• Canada western hard white spring (CWHWS)
• Canada western amber durum (CWAD)
• Canada western soft white spring (CWSWS)
• Canada western red winter (CWRW)
• Canada western extra strong (CWES)
• Canada prairie spring white (CPSW)
• Canada prairie spring red (CPSR)

4
Introduction

• Wheat harvesting 13 to 15 m.c. (normally ? 15
  m.c.) drying storage
• Wheat _at_ 12 to 13 m.c.
• - safe moisture for effective storage
• - prevention of spoilage by fungi
• - sprouting before processing can be prevented
• Wheat class identification Major task in grain
  handling facilities
• Visual method (common method)
• - to identify different wheat classes
• - but not to identify their moisture levels
• Machine vision, PAGE, and HPLC methods

5
Introduction

• Near infrared (NIR) hyperspectral imaging
• - Machine vision NIR spectroscopy
• - to develop a rapid and consistent method
• - Non destructive, non subjective method
• - Food science, Chemistry, Pharmaceuticals,
  Animal science
• - Grain storage wheat class identification,
  moisture identification, protein and oil
  content determination in wheat

6
Objectives

• To identify western Canadian wheat classes at
  different moisture levels by developing
  statistical classification models

7
Materials and Methods

• Hyperspectral imaging system

1. Bulk wheat sample, 2. Liquid crystal tunable
filter (LCTF), 3. Lens, 4. NIR camera, 5. Copy
stand, 6. Illumination, and 7. Data processing
system.
8
Methods and Materials

• Wheat classes CWRS, CWSWS, CWHWS, CWRW, and CWES
• Moisture levels 12, 14, 16, 18, and 20
• 100 images/class/m.c. 960 to 1700 nm 10 nm
  interval

9
Methods and Materials

• Relative reflectance intensity, R (S-D/W-D
• where
• R relative reflectance intensity of each
  slice of the NIR hyperspectral image of wheat
• S reflectance intensity of each slice of
  the NIR hyperspectral image
• D reflectance intensity of the dark current
• W reflectance intensity of a 99 reflectance
  standard white panel
• Linear and quadratic discriminant analyses
  statistical classification models

10
Results

• Linear discriminant analysis

11
Results

• Quadratic discriminant analysis

12
Results

• Top 10 wavelengths in wheat class identification
• No. Wavelength (nm) Partial R2 ASCC
• 1 1310 0.66 0.03
• 2 1450 0.80 0.06
• 3 1060 0.76 0.09
• 4 1700 0.72 0.12
• 5 1330 0.55 0.13
• 6 1200 0.33 0.14
• 7 1160 0.33 0.15
• 8 1090 0.29 0.16
• 9 1490 0.28 0.16
• 10 1070 0.26 0.18
• ASCC Average squared canonical correlation

13
Discussion

• Identification of waxy wheat 1 to 10 principal
  component scores as input 42 to 71 (LDA) and
  46 to 71 (QDA) (Delwiche and Graybosch 2002)
• Classification of barley based on ergosterol
  levels - ? 86.6 (LDA and QDA) (Balasubramanian
  et al. 2006)
• Mohan et al. 2005 Mean classification accuracies
  89.1 (LDA, Top 2 Ref. features), 99.1 (LDA,
  Top 5 Ref. features) Cereal grains
  classification

14
Discussion

• 81 100 (LDA) and 60 89 (QDA) relative
  reflectance intensities Identification of wheat
  classes at different moisture levels

15
Conclusions and future work

• NIR hyperspectral imaging was found useful to
  identify different moisture level wheat classes
  with the extracted relative reflectance
  intensities as input for classification
• This technique could be used to develop an
  automatic grain assessment tool
• Wheat samples from different crop years and
  locations could be included in the sample space
  to improve the robustness and classification
  efficiency of the models

16
Acknowledgements

• Dr. Digvir S. Jayas
• Dr. Jitendra Paliwal
• Dr. Noel D.G. White