In this study, a hyperspectral imaging system of 866.4–1701.0 nm was selected and combined with multivariate methods to identify wheat kernels with different concentrations of omethoate on the surface. In order to obtain the optimal model combination, three preprocessing methods (standard normal variate (SNV), Savitzky–Golay first derivative (SG1), and multivariate scatter correction (MSC)), three feature extraction algorithms (successive projections algorithm (SPA), random frog (RF), and neighborhood component analysis (NCA)), and three classifier models (decision tree (DT), k-nearest neighbor (KNN), and support vector machine (SVM)) were applied to make a comparison. Firstly, based on the full wavelengths modeling analysis, it was found that the spectral data after MSC processing performed best in the three classifier models. Secondly, three feature extraction algorithms were used to extract the feature wavelength of MSC processed data and based on feature wavelengths modeling analysis. As a result, the MSC–NCA–SVM model performed best and was selected as the best model. Finally, in order to verify the reliability of the selected model, the hyperspectral image was substituted into the MSC–NCA–SVM model and the object-wise method was used to visualize the image classification. The overall classification accuracy of the four types of wheat kernels reached 98.75%, which indicates that the selected model is reliable.
Novel segmented stacked autoencoder for effective dimensionality reduction and feature extraction in hyperspectral imaging