Detection of Sprouted and Midge- Damaged Wheat kernels Using Near-Infrared Hyperspectral Imaging C.B. Singh, D.S. Jayas, J. Paliwal, N.D.G. White CSBE\ SCGAB Annual Conference Vancouver, BC July 13-16, /16/08CSBE 2008 Conference
Brief overview 07/16/08CSBE 2008 Conference Sprout and midge-damage Current detection methods Hyperspectral imaging Calibration and pre-processing Data reduction and feature selection Classification results Conclusion
Sprout damage Sprouting is considered one of the important degrading factors of wheat in Canada Results in decrease in starch, increase in sugar, dry matter loss, and poor bread making quality Sprouted kernels are more vulnerable to insect infestation 07/16/08CSBE 2008 Conference
Midge-damage Occurs when the orange wheat blossom midge larva feeds on the developing wheat kernels Causes the kernels to shrivel, crack, and become deformed and misshapen Splits the kernel’s pericarp, facilitating the water uptake and hence sprouting in poor weather 07/16/08CSBE 2008 Conference
07/16/08CSBE 2008 Conference Visual inspection Falling number test X-rays NIR spectroscopy Detection methods
07/16/08CSBE 2008 Conference To identify the spectral bands which are appropriate for detection of sprout and midge-damage in wheat To develop calibration algorithms to detect sprouted and midge-damaged wheat kernels Research objectives
Sample preparation Artificially sprouted and midge-damaged wheat samples Five locations: Camrose East, AB; Vegreville, AB; North Battleford, SK; Cutknife, SK; and Yorkton, SK Midge-damaged kernels were manually selected 07/16/08CSBE 2008 Conference
Imaging system 07/16/08CSBE 2008 Conference Stand InGaAs camera LCTF Area of view Halogen-tungsten lamp Data storage and analysis
Calibration & pre-processing Dark current measurement Co-addition Filtering and segmentation Transformation into reflectance 07/16/08CSBE 2008 Conference
Hypercube Reflectance Wavelength x y 07/16/08CSBE 2008 Conference labeling (Singh et al. 2007)
07/16/08CSBE 2008 Conference PC loadings
07/16/08CSBE 2008 Conference Reflectance spectra
07/16/08CSBE 2008 Conference First PC score images HealthySproutedMidge-damaged
Wavelengths , , and nm were identified as significant wavelengths Features: Mean, maximum, minimum, median, standard deviation, and variance Discriminant analysis (linear, quadratic, and Mahalanobis) were used for classification 07/16/08CSBE 2008 Conference Classification model
07/16/08CSBE 2008 Conference Classification results ClassifierClassification accuracy(%) HealthySprouted Linear Quadratic Mahalanobis Table: Classification of sprouted wheat kernels
07/16/08CSBE 2008 Conference Classification results Actual classClassification accuracies by various discriminant classifiers (%) LinearQuadraticMahalanobis Healthy Yorkton Camrose East Cutknife Vegreville North Battleford Table: Midge-damaged kernels (Crease-down)
07/16/08CSBE 2008 Conference Classification results Actual classClassification accuracies by various discriminant classifiers (%) LinearQuadraticMahalanobis Healthy Yorkton Camrose East Cutknife Vegreville North Battleford Table: Midge-damaged kernels (Crease-up)
Three wavelengths , , and nm were found as the most significant Linear discriminant classifier classified more than 96% healthy kernels Mahalanobis discriminant classifier correctly classified % damaged kernels 07/16/08CSBE 2008 Conference Conclusion
Mr. Ian Wise 07/16/08CSBE 2008 Conference Acknowledgements
07/16/08CSBE 2008 Conference Thank You!