Rapid nondestructive detecting of wheat varieties and mixing ratio by combining hyperspectral imaging and ensemble learning

Wheat is the main raw material for brewing Chinese liquor, and differences in the wheat varieties and mixing ratio will affect its quality and flavor. In this study, hyperspectral imaging (HSI) was combined with ensemble learning models for the classification and determination of the mixing ratio of wheat. The spectral information and textural and shape features of each wheat grain were respectively extracted. The density-based spatial clustering of applications with noise (DBSCAN) algorithm was used to remove abnormal data, and Savitzky-Golay combined multiplicative scatter correction (SG-MSC) was used to pre-process the spectra of the wheat samples. The characteristic wavelengths were then extracted using the competitive adaptive reweighted sampling (CARS) algorithm, and the classification effects of BP-Adaboost models were compared when using feature spectral data, image features, and fusion data as the input. The recognition effects and visualization of the validation set proved the optimal classification of feature spectral data fused with shape features; the average accuracy was 92.29% and the maximum deviation range of mixing ratio prediction was 5%. With the addition of wheat classification categories, this method still achieved excellent results. The results prove the feasibility of using fusion data with HSI combined with ensemble learning models for the classification and mixing ratio detection of wheat.