Identification of rice-weevil (Sitophilus oryzae L.) damaged wheat kernels using multi-angle NIR hyperspectral data

Rice-weevil (RW) has great harm to wheat kernels, when the RW-damaged wheat kernels are mixed into the sound wheat kernels, the quality of wheat products will be seriously reduced. In this work, a novel method of calibrating the model based on multi-angle near-infrared(NIR) hyperspectral data was proposed to identify sound wheat kernels and RW-damaged wheat kernels. Hyperspectral images on four sides of each wheat kernel were collected. Some commonly used spectral preprocessing methods and two feature extraction algorithms (successive projections algorithm (SPA), and random frog (RF)) were used to combine three machine learning models (partial least squares discriminant analysis (PLS-DA), linear discriminant analysis (LDA), and support vector machine (SVM))) for modeling analysis. After multivariate data analysis, it was found that standard normal variate (SNV)-SPA-LDA was the best hybrid model. Finally, considering the actual situation, the reliability of the calibrated model was verified by external validation, and the classification results were visualized, in which the accuracy, sensitivity and specificity of the model were 97%, 98% and 96%, respectively. The results indicated that the model calibrated by hyperspectral data from four sides of wheat kernel was reliable.

Automated summary

This study proposed a novel method to identify sound wheat kernels and RW-damaged wheat kernels using multi-angle near-infrared hyperspectral data calibration model. The best hybrid model SNV-SPA-LDA was found through multivariate data analysis, with an accuracy of 97%, sensitivity of 98%, and specificity of 96%. The results indicated the reliability of the calibrated model based on hyperspectral data from four sides of wheat kernel.