高光谱成像
共线性
主成分分析
偏最小二乘回归
氮气
光谱带
数学
冗余(工程)
天蓬
生物系统
化学
遥感
环境科学
植物
计算机科学
统计
人工智能
生物
操作系统
地质学
有机化学
作者
Jingang Wang,Tian Tian,Haijiang Wang,Jing Cui,Yongqi Zhu,Wenxu Zhang,Xuanmeng Tong,Tianhang Zhou,Zhenkang Yang,Jiaqi Sun
标识
DOI:10.1016/j.compag.2021.106390
摘要
• Leaf nitrogen concentration was significantly correlated with MDA and POD activities. • Band extraction by SPA reduced the collinearity and redundancy of N-sensitive bands. • Combining of LNC and OA-sensitive bands improved the model accuracy. Oxidase activities (OA) are highly correlated with the nitrogen concentration in crop leaves. To improve the universality and inter-annual repeatability of the model for estimating cotton leaf nitrogen concentration (LNC), a method for model construction was proposed based on the combination of the bands sensitive to LNC with the bands sensitive to OA. In this plot experiment, 320 and 250 sets of hyperspectral data of cotton leaves in seedling stage, bud stage, initial flowering stage, full flowering stage, and boll setting stage were collected in 2019 and 2020, respectively by using hyperspectral technology, and the LNC and OA were also measured in indoor biochemical experiments. Then, successive projection algorithm (SPA) was used to analyze the LNC and OA-sensitive bands in the original spectrum and five kinds of spectral conversions in 2019, to construct the partial least squares regression (PLSR) and principal component regression (PCR) models. Finally, the accuracy of the models were verified using the spectral data in 2020. The results showed that the selection of LNC and OA-sensitive bands could greatly reduce the collinearity and redundant information among bands. The accuracy of the models based on the LNC and OA-sensitive bands in all stages were higher than those of the models based on the LNC-sensitive bands. The optimal was the model based on the malondialdehyde (MDA), peroxidase (POD), and LNC sensitive bands in full flowering stage, with determination coefficients (R 2 ) of 0.846, root mean squared error (RMSE) of 3.081, and residual prediction deviation (RPD) of 2.975. The universality and inter-annual repeatability of the optimal model were significantly improved, with R 2 increasing by 12.24%-79.89% and RMSE reducing by 19.80%-72.52%, compared with those of the model based on the LNC-sensitive bands. Besides, the accuracy and stability of PLSR models were significantly higher than those of PCR models. In conclusion, the combination of LNC and OA-sensitive bands could obviously improve the accuracy and universality of the LNC estimation model. This study provides a new method for improving the accuracy and universality of crop nitrogen estimation model.
科研通智能强力驱动
Strongly Powered by AbleSci AI