Nondestructive Rapid Identification of Soybean Varieties Using Hyperspectral Imaging Technology (In Engl.)

Abstract

Hyperspectral imaging technology was used to classify four types of soybean varieties. The reflectance spectra of four varieties of soybeans were extracted from hyperspectral images covering wavelengths from 400 to 1000 nm. Firstly, exploratory principal component analysis and linear discriminant analysis (LDA) were carried out to infer the separability of soybean spectral data. Secondly, the spectral data were preprocessed using multiplicative scattering correction (MSC), Savitzky–Golay (SG) smoothing, and MSC and SG smoothing together. Finally, classification models based on LDA, support vector machine (SVM), and k nearest neighbor (KNN) were established based on the full wavelengths or feature wavelengths. MSC and SG smoothing joint preprocessing of the spectral data was applied to establish the SVM classification model based on the full wavelengths, which returned a classification accuracy of 95.19%. Random forest was used to select the feature wavelengths from the full wavelengths to establish the LDA classification model, and the classification accuracy reached 82.69%. The results showed that the hyperspectral imaging technique combined with SVM, KNN, and LDA algorithms can be used to classify different soybean varieties in a fast and nondestructive way.