Determination of damaged wheat kernels with hyperspectral imaging analysis

Hyperspectral imaging was applied to classify the damaged wheat kernels and healthy kernels. The spectral information was extracted from damaged wheat kernels and healthy kernels samples. The effective wavelengths were obtained from spectral of 865-1711 nm by X-loadings of principal component analysis (PCA) and successive projection algorithm (SPA) method, respectively. Partial least square method (PLS) and least square-support vector machine (LS-SVM) were then used to build classification models on full spectral data and effective wavelengths dataset, respectively. The results showed that the classification accuracy of every LS-SVM model was the best, being 100%. While the accuracy of the PLS model was slightly lower, still over 97%. The confusion matrix showed that several damaged wheat kernels samples were misclassified as healthy samples, while all healthy samples were correctly classified. The overall results indicated that hyperspectral imaging could be used for discriminating the damaged wheat kernels and could provide a reference for detecting other grain kernels grading degrees. Further, this study can provide a research basis for the development of online or portable detectors on grain damaged kernels recognition, which will be beneficial for grain grading or post-harvest quality processing of other grains.