Vigor identification of maize seeds by using hyperspectral imaging combined with multivariate data analysis

As an important food crop in China, rapid and accurate discrimination of maize seed vigor is crucial for agricultural development. In this study, 1,680 maize seeds were subjected to artificial accelerated aging treatment with "Zhengdan 958″ as the research object to verify the difference in vigor between damaged and healthy seeds based on the results of the standard germination tests. A Hyperspectral imaging (HSI) system was used to obtain spectral information of samples, and the original spectra were preprocessed using second-order Savitzky-Golay smoothing (SG-2), first derivative (FD), detrending (DE), standard normal variate (SNV), and multiplicative scatter correction (MSC) methods. The 49, 60, 54, and 40 numbers of feature wavelengths (nm) were extracted from the processed spectra using successive projections algorithm (SPA), uninformative variable elimination (UVE), interval random frog (IRF), and iteratively variable subset optimization (IVSO), respectively. The decision tree (DT), support vector machine (SVM), K-nearest neighbor (KNN), linear discriminant analysis (LDA), random forest (RF), and artificial neural network (ANN) models were constructed based on full wavelength and feature wavelength, of which the best model was DE-UVE-ANN, and its identification accuracy reached 95.24 %. The experimental results show that the UVE algorithm is the most effective method for preprocessing, and the accuracy of LDA and ANN models built based on it is above 85.71 % and 89.76 %, respectively. In addition, the hyperspectral images were visualized based on an object-oriented approach to observe intuitive identification results. The results indicate that the proposed method is instructive for the vigor identification of maize seeds.