Please use this identifier to cite or link to this item:
|Title:||NIR Imaging and its Application to Wheat Grading|
|Keywords:||NIR;NIR image;wheat;wheat grading|
|Abstract:||Two topics related to near infrared (NIR) imaging technology are studied in this thesis. The first is on the calibration of line-scan NIR imaging systems, the second covers the feasibility of applying the NIR imaging technology for wheat grading. In the first study, a methodology is proposed to pretreat the NIR image data acquired by the line-scan NIR imaging system used in this thesis to reduce the systematic noise introduced by the imaging system. This calibration in a standardization methodology is shown to improve the result of multivariate image analysis (MIA) based on multi-way principal component analysis (MPCA). This method represents a practical and easily used tool for calibration of line-scan NIR imaging systems in that it does not employ expensive standard reflectance material. In the second study, two projects are accomplished. In the first project, NIR imaging is used to classify different classes of wheat kernels. Multivariate statistical algorithms, soft independent modeling of class analogy (SIMCA) and partial least square discriminant analysis (PLS-DA) are used to discriminate between different types of wheat kernels using spectral features from NIR images. A new strategy of implementing multiclass PLS-DA algorithm is proposed in this part. The results from this study show that NIR imaging provides a potentially fast and objective method for qualitatively evaluating certain characteristics of wheat samples, such as fungal infection, sprout damage and foreign types of grain, which are now graded manually in wheat industry. In the second project, NIR imaging is used to predict the "falling number" (FN) of wheat samples. Three models are built between the features extracted from NIR images of the wheat kernels and the falling number measurements made on bulk samples. One uses a regular PLS algorithm, one uses the orthogonal partial least square (O-PLS) algorithm and the other uses the PLS plus canonical correlation analysis (PLS+CCA) algorithm. The models are analyzed and the performance of the algorithms is discussed. The errors in the prediction of the O-PLS model are investigated. The results from this study indicate that NIR imaging is a promising method for the rapid assessment of the FN of wheat samples.|
|Appears in Collections:||Digitized Open Access Dissertations and Theses|
Files in This Item:
|liu_zheng_2006Mar_masters.pdf||18.74 MB||Adobe PDF||View/Open|
Items in MacSphere are protected by copyright, with all rights reserved, unless otherwise indicated.