基于无人机高光谱遥感的冬小麦全氮含量反演

氮素是作物生长发育必需的营养元素之一，作物的全氮含量是表征其氮素状况的主要指标。田块尺度的冬小麦全氮含量空间分布监测可以辅助其精准定量追肥，减少环境污染。无人机高光谱遥感具有分辨率高、时效性高、成本低等优势，可为作物长势信息反演提供重要数据源。XGBoost（extreme gradient boosting）作为一种新兴集成学习算法，运行效率高，泛化能力强，可以有效的应用于构建冬小麦全氮含量遥感反演模型，预测田块尺度冬小麦全氮含量空间分布。以农业部蒙城砂姜黑土生态环境站内拔节期冬小麦为研究对象，开展以下工作: （1）以低空无人机搭载高光谱成像仪获取冬小麦拔节期冠层成像光谱影像，结合地面采样数据，获取126个样点全氮含量数据；（2）分析拔节期冬小麦冠层光谱特征，并根据Person相关系数分析176个波段的光谱反射率与全氮含量之间的相关性；（3）构建基于XGBoost算法的不同土壤肥力条件下拔节期冬小麦全氮含量无人机高光谱反演模型。结果表明: （1）176个波段（400~1 000 nm）的光谱反射率与冬小麦全氮含量之间具有较强的相关性，除了735.5 nm外其他波段光谱反射率与全氮含量之间的相关系数均大于0.5；（2）基于XGBoost算法构建的拔节期冬小麦全氮含量无人机高光谱遥感反演模型具有较高的反演精度（R2=0.76，RMSE=2.68）；（3）基于XGBoost算法的冬小麦全氮含量反演模型可以获取不同土壤肥力条件下田块尺度的全氮含量空间分布图，总体上呈现较为显著的空间差异。该研究可为冬小麦精准定量追肥提供一定的科学依据，也为发展无人机高光谱遥感的精准农业应用提供了参考。

Winter Wheat Total Nitrogen Content Estimation Based on UAV Hyperspectral Remote Sensing

Nitrogen is one of the necessary nutrient elements for crops’ growth and development, and crops’ total nitrogen content is the main index to characterize its nitrogen status. Monitoring the spatial distribution of winter wheat total nitrogen content at the field scale can assist in accurate and quantitative topdressing and reduce environmental pollution. UAV (Unmanned aerial vehicle) hyperspectral data can provide an important data source for crop growth information inversion due to its high resolution, high timeliness and low cost. XGBoost (Extreme Gradient Boosting), a new ensemble learning algorithm with high efficiency and strong generalization ability, can be effectively applied to build a winter wheat total nitrogen content estimation model based on remote sensing data and predict the spatial distribution of winter wheat total nitrogen content at field scale. Therefore, this study selected the winter wheat at the jointing stage in the national soil quality observation and experimental station as the study object and carried out the following work: (1) we obtained the canopy imaging spectral image of winter wheat at the jointing stage with a hyperspectral imager mounted on a low-altitude UAV, and total nitrogen content data of 126 samples combined with ground sampling data. (2) The spectral characteristics of the winter wheat canopy at the jointing stage were analyzed, and the correlation between spectral reflectance of 176 bands and total nitrogen content was analyzed according to the Person correlation coefficient. (3) A winter wheat total nitrogen content estimation model based on UAV hyperspectral at the jointing stage was built with the XGBoost algorithm under different soil fertility conditions. The results showed that: (1) there was a strong correlation between spectral reflectance and total nitrogen content of winter wheat in 176 bands, and the correlation coefficients between spectral reflectance and total nitrogen content in all bands except 735.5 nm were greater than 0.5; (2) The UAV hyperspectral remote sensing estimation model of winter wheat total nitrogen content at jointing stage based on XGBoost algorithm shows high accuracy (R2=0.76, RMSE=2.68); (3) The estimation model of winter wheat total nitrogen content based on XGBoost algorithm can obtain the spatial distribution map of total nitrogen content at field scale under different soil fertility conditions, which shows a significant spatial difference on the whole. This study can provide a scientific basis for the accurate and quantitative topdressing of winter wheat and also provide a reference for the application of UAV hyperspectral remote sensing in precision agriculture.