结合SPA和PLS法提高冬小麦冠层全氮高光谱估算的精确度

【目的】 冠层高光谱全波段信息可以在小麦拔节期快速无损地估算叶片的氮含量。本研究结合连续投影算法 (SPA) 和偏最小二乘 (PLS) 技术，筛选了冬小麦拔节期冠层光谱对叶片氮含量的敏感特征波段，以期为冬小麦关键生育期氮素含量的遥感估算提供理论依据和技术支持。 【方法】 以陕西关中地区2015—2016年冬小麦小区试验为基础，基于连续投影算法 (SPA) 提取冬小麦叶片全氮含量的冠层光谱敏感波段，并结合偏最小二乘 (PLS) 回归法建立基于敏感特征波段的冬小麦拔节期叶片氮含量估算模型。 【结果】 SPA算法从冬小麦338～2510 nm的冠层光谱中优选出了1985 nm、2474 nm、1751 nm、1916 nm、2507 nm、1955 nm、2465 nm和344 nm共计8个叶片全氮含量的敏感特征波段，波段数目下降了98.9%，有效降低了光谱信息的冗余；基于敏感特征波段构建的叶片氮含量偏最小二乘回归模型的决定系数和均方根误差分别为0.82和0.28，模型验证方程的决定系数和均方根误差分别为0.84和0.21，模型的相对预测偏差大于2，具有较高的精度和良好的预测能力。 【结论】 与常用植被指数的叶片氮含量估算模型相比，连续投影算法 (SPA) 结合偏最小二乘 (PLS) 方法的叶片氮含量估算精度更高，稳定性更强，可以作为冬小麦拔节期叶片氮含量的高光谱估算方法。

Increasing accuracy of hyper-spectral remote sensing for total nitrogen of winter wheat canopy by use of SPA and PLS methods

【Objectives】 Canopy hyper-spectral reflectance has been used for the estimation of nitrogen in the key growing stages of crop rapidly and non-destructively. In this paper, the successive project algorithm (SPA) and the partial least squares (PLS) were used to help screening more sensitive wave bands, so as to increase the accuracy of hyper-spectral remote sensing in the estimation of nitrogen contents in winter wheat at jointing stage. 【Methods】 Based on a plot experiment of winter wheat in Guanzhong area of Shaanxi Province in 2015–2016, the canopy spectrum sensitive wave bands of leaf nitrogen contents of winter wheat were calculated by SPA. The leaf nitrogen contents estimation model of winter wheat based on the sensitive feature bands at the jointing stage was established by using PLS regression method. 【Results】 Through the SPA, 8 sensitive feature wave bands of leaf nitrogen contents were selected in canopy spectrum ranging from 338 nm to 2510 nm of winter wheat, including 1985 nm, 2474 nm, 1751 nm, 1916 nm, 2507 nm, 1955 nm, 2465 nm and 344 nm. The result decreased the number of bands by 98.9%, and effectively reduced the redundancy of spectral information. The determination coefficient and root-mean-square error of the leaf nitrogen contents regression model that was established by PLS based on the sensitive feature bands were 0.82 and 0.28, respectively. The determination coefficient of the verification equation of model was 0.84, the root-mean-square error was 0.21 and the residual prediction deviation was large than 2, which indicated that the model had high precision and good prediction ability. 【Conclusions】 Compared with the leaf nitrogen contents estimation model developed using common vegetation indices, the SPA combining with PLS has higher precision and more stability, which can be used as the hyper-spectral reflectance estimation method for the leaf nitrogen contents of winter wheat at the jointing stage.