Nondestructive Testing of Sunflower Seed Vigor Based on Hyperspectral Images and Chemometrics

Abstract

Sunflower is an important crop, and the vitality and moisture content of sunflower seeds have an important influence on their planting and yield. Therefore, it’s of great significance for agricultural production to detect the vigor and moisture content of sunflower seeds quickly and accurately. This study aims to study and predict the vitality and moisture content of sunflower seeds by analyzing the spectral characteristics of sunflower seeds in the wavelength range of 384-1034nm using hyperspectral technology combined with the moisture content detection results of sunflower seeds. In this study, the hyperspectral data were first preprocessed by Savitzky-Golay smoothing, standard normal variable (SNV) or multiple scattering correction (MSC) to reduce the impact of baseline drift and tilt, ensuring the accuracy and reliability of the data. Subsequently, principal component analysis (PCA), extreme gradient boosting (XGBoost), and stacked autoencoders (SAE) were used for feature band extraction to enhance the interpretability and predictive ability of the data. During the modeling process, various commonly used data analysis algorithms were employed, including random forest (RF), LightGBM, StackingClassifier, and StackingRegressor, to model the viability classification and moisture content of sunflower seeds, respectively. After experimentation and comparison, the best model selection was determined to be the SG-SAE-StackingClassifier model, which achieved a prediction accuracy of 97.30% for the classification of sunflower seed viability. Additionally, the SG-XGBoost-StackingRegressor model achieved a determination coefficient ([[EQUATION]]) value of 0.9758 and a root mean square error (RMSE) value of 0.7644 for predicting the moisture content of sunflower seeds. The comprehensive experimental results indicate that the combination of hyperspectral technology with multiple data analysis algorithms can rapidly and accurately predict the viability and moisture content of sunflower seeds. This provides important references and support for the quality assessment of sunflower seeds and agricultural production. This research provides a beneficial approach and method for utilizing hyperspectral technology to correlate the external manifestations of effective vitality in sunflower seeds. It holds important theoretical and practical significance.

Keywords: hyperspectral imaging technology, Sunflower seed vitality, Seed moisture content, Feature extraction, Ensemble learning model