Polyethylene glycol (PEG) is one of the most commonly used seed initiators, in order to regulate the seed water uptake during seed priming. This study aims to investigate the effect of seed water uptake on the priming behavior under different PEG concentrations and priming times using low-field nuclear magnetic resonance (LF-NMR) technology. The aqueous phase state of maize seeds was dynamically monitored to be initiated by the different PEG concentrations. A prediction model of maize seed germination was constructed to clarify the relationship between seed water uptake and priming effect using machine learning. LF-NMR data was collected from the maize seeds at each node of priming time. A systematic analysis was carried out to determine the effects of PEG priming on seed germination and seedling growth. The optimal conditions of priming treatment were determined to evaluate the physical and chemical indicators, such as the vigor index of maize seeds. The pattern of water uptake was obtained in the maize seeds during priming. A regression model was also constructed to predict the priming effect of maize seeds. The main findings were summarized as follows: Firstly, there were some effects of PEG priming on the germination and seedling growth of maize seeds. Two-way ANOVA analysis showed that there were highly significant effects of PEG priming time and priming concentration on the germination indexes of maize seeds. Among them, the effect of priming time on the germination indexes of maize was much larger than that of priming concentration. With the increase of priming time (0-48 h), the viability of maize seeds showed a tendency to increase and then decrease, where the most suitable priming time was 16 h. Secondly, the water uptake of maize seed was elucidated during priming. The internal water of maize seeds was divided into two kinds of water components: bound water (0.1 ms < T₂₁ < 20 ms) and free water (20 ms < T₂₂ < 400 ms), according to the length of LF-NMR T₂ relaxation time. With the increase of priming time (0-48 h), the content of bound water increased and then gradually stabilized, while the content of free water continued to rise. The total water content increased and then gradually slowed down. The content of bound water had just entered the stagnation stage, while the increasing trend of A₂₁ signal amplitude decreased basically without increase, indicating an appropriate indicator to judge the priming time of maize seeds. Finally, the medium Gaussian support vector machine (SVM) model and exponential gaussian process regression (GPR) model were constructed to assess the priming effect of maize seeds, respectively, according to the data before and after the screening of LF-NMR feature parameters. Among them, the exponential GPR model was achieved in the validation set R² of 0.920 for maize seed germination after feature screening, which was better than 0.900 before that. The water absorption kinetics of the seed was elucidated to verify the priming efficacy. The experiment proved that the moisture variations during seed priming detected by LF-NMR can be expected for the rapid screening of the optimal priming conditions. The finding can also provide a new idea for the parameters setting and rapid evaluation during corn seed priming.

Aging and deterioration of corn seeds have been two of the key influencing factors on their vitality. Therefore, it is of great significance to identify the storage time of seeds for the preservation of germplasm resources and the quality identification of seeds. This study aims to identify the stored corn seeds over the different years using machine learning. 103 corn varieties were also selected for the correlation analysis among the physicochemical parameters and hyperspectral bands. Hyperspectral imaging was employed to obtain the spectral data from both sides of the seed embryonic and endosperm. The near-infrared component analyzer (IM9500) and portable ultraviolet-visible fluorescence spectrometer (Multi-plex) were utilized to acquire the physicochemical parameters of corn seeds. Variance and linear discriminant analysis was conducted to determine the trends of physicochemical parameters within corn seeds, and their impact on the discrimination of storage time. The influencing factors were selected with the physicochemical parameters on the discrimination of corn seed storage time, as the storage time increased. The hyperspectral data was also performed on the black-and-white correction. The threshold segmentation was used to effectively separate the background area, in order to obtain the region of interest (ROI). Among them, the spectral average of all pixel points on the image served as the spectral data of ROI. Five preprocessing methods, including savitzky-golay (SG) smoothing, standard normal variate transformation (SNV), multiplicative scattering correction (MSC), first derivative (1-Der), and second derivative (2-Der), were applied to the spectral data, in order to eliminate the interference signals during spectral acquisition, such as the background noise, baseline drift, and stray light. Feature wavelength was selected from the complete spectral data, due mainly to the hyperspectral images with many bands and redundant information. Competitive adaptive reweighted sampling (CARS) and uninformative variable elimination (UVE) were used to select the feature wavelength. Support vector machine (SVM), back propagation neural network (BPNN), and convolutional neural network (CNN) classification models were developed to preprocess the spectral data from the embryo and endosperm surfaces. The spectral data was obtained as the input for the models. The classification data was compared with the different models. The results indicated that better discrimination was achieved in the spectral data from the embryonic side on the storage time of corn seeds. Modeling with the preprocessed and feature-selected spectral data significantly outperformed that with the raw data. Pearson correlation analysis was conducted among physicochemical parameters that significantly affected the seed age discrimination and hyperspectral bands. These parameters were selected with the highly correlated feature bands for modeling. The comparison was made on the modeling accuracy of feature wavelength selection, correlation coefficient, and the combination of correlation and feature wavelength selection. The performance of the model was verified to detect across multiple corn varieties, indicating the high generalization. BPNN classification model with the correlation coefficient and feature wavelength shared the highest accuracy, with a single kernel prediction accuracy of 92.3% and a Colony prediction accuracy of 94.4%. This finding can provide a significant theoretical basis and practical implications for the precise management of corn seed storage in the seed industry.