Detection of peanut seed vigor based on hyperspectral imaging and chemometrics

• DOI: 10.3389/fpls.2023.1127108
• Publication Date: 2023-02-27T06:47:50Z
• Authors: Zhiyong Zou; Jie Chen; Weijia Wu; Jinghao Luo; Tao Long; Qingsong Wu; Qianlong Wang; Jiangbo Zhen; Yongpeng Zhao; Yuchao Wang; Yongming Chen; Man Zhou; Lijia Xu
• Keywords: Chemometrics

Rapid nondestructive testing of peanut seed vigor is great significance in current research. Before seeds are sown, effective screening high-quality for planting crucial to improve the quality crop yield, and vitality one important indicators evaluate quality, which can represent potential ability germinate quickly whole grow into normal seedlings or plants. Meanwhile, advantage technology that themselves will not be damaged. In this study, hyperspectral superoxide dismutase activity were used detect vigor. To investigate predict activity, spectral characteristics wavelength range 400-1000 nm analyzed. The data processed by a variety hot spot algorithms. Spectral preprocessed with Savitzky-Golay (SG), multivariate scatter correction (MSC), median filtering (MF), effectively reduce effects baseline drift tilt. CatBoost Gradient Boosted Decision Tree feature band extraction, top five weights characteristic bands classification 425.48nm, 930.8nm, 965.32nm, 984.0nm, 994.7nm. XGBoost, LightGBM, Support Vector Machine Random Forest modeling classification. XGBoost partial least squares regression establish value model. results indicated MF-CatBoost-LightGBM was best model classification, accuracy result 90.83%. MSC-CatBoost-PLSR optimal value. show R2 0.9787 RMSE 0.0566. suggested could correlate external manifestation