当前位置：
X-MOL 学术
›
Infrared Phys. Technol.
›
论文详情
Our official English website, www.x-mol.net, welcomes your feedback! (Note: you will need to create a separate account there.)
Rapid estimation of leaf nitrogen content in apple-trees based on canopy hyperspectral reflectance using multivariate methods
Infrared Physics & Technology ( IF 3.3 ) Pub Date : 2020-12-01 , DOI: 10.1016/j.infrared.2020.103542 Shaomin Chen , Tiantian Hu , Lihua Luo , Qiong He , Shaowu Zhang , Mengyue Li , Xiaolu Cui , Hongxiang Li
Infrared Physics & Technology ( IF 3.3 ) Pub Date : 2020-12-01 , DOI: 10.1016/j.infrared.2020.103542 Shaomin Chen , Tiantian Hu , Lihua Luo , Qiong He , Shaowu Zhang , Mengyue Li , Xiaolu Cui , Hongxiang Li
Abstract Rapid estimation of crop nitrogen (N) status is helpful for achieving precise N fertilizer management, so as to promote the collaborative improvement of yield, quality and N use efficiency. To establish a method for rapidly estimating N status of apple-trees based on hyperspectral remote sensing, a field experiment on apple trees with four levels of N application were conducted in Luochuan, Yan'an, China in 2018 and 2019. Hyperspectral data at apple tree canopy scale were obtained at different phenological stages. The outliers in the dataset were detected by Monte-Carlo cross-validation (MCCV). A total of 15 forms of spectral data were analyzed, including the raw spectrum (RS), Savitzky-Golay (SG) smoothing, normalization by the mean (NME), standard normal transformation (SNV), multiplicative scatter correction (MSC) and their combination with the first-order derivative (FD) or second-order derivative (SD). The results showed that SNV-FD was the best preprocessing method for apple tree canopy spectral data on the Loess Plateau. For SNV-FD data, competitive adaptive reweighted sampling (CARS), successive projection algorithm (SPA), random frog (Rfrog), and partial least squares (PLS) were used to extract feature variables. The feature wavelengths extracted by CARS, SPA and Rfrog were widely distributed in visible and near-infrared (VIS/NIR) range. The feature variables extracted by PLS were fewer, but had a strong interpretation. Based on these feature variables, linear and nonlinear models including partial least squares regression (PLSR), support vector machine (SVM), back-propagation artificial neural network (BPANN), extreme learning machine (ELM) and random forest (RF), were used to establish the prediction model. Among all the models, four nonlinear modeling methods were superior to the linear method. The model by Rfrog-ELM achieved the best results (R2P = 0.843, RMSEP = 2.461 g·kg−1, RPD = 2.508). This study demonstrated that the combination of SNV-FD, Rfrog and ELM was feasible and reliable for estimating the leaf N content in apple-trees on the Loess Plateau, China.
更新日期：2020-12-01
京公网安备 11010802027423号