基于无人机多源遥感信息融合的冬小麦叶片氮素含量监测方法

Monitoring Method for Winter Wheat Leaf Nitrogen Content Based on UAV Multi-source Remote Sensing Fusion

摘要:

实时监测氮素养分状态对于提高肥料效率和作物产量、品质至关重要。作物氮素监测在精准农业中具有重要意义,但传统的土壤和叶片采样方法难以满足大范围、实时监测的需求。本研究结合RGB与多光谱传感器的多源数据,构建了冬小麦叶片氮含量(Leaf nitrogen content,LNC)估算模型,旨在提升氮素监测精度。首先基于无人机 RGB 与多光谱影像提取冬小麦的多源特征信息,并通过结合Pearson相关性分析与变量投影重要性( Variable importance in projection,VIP)的“两段式”特征筛选策略识别关键变量,融合颜色空间、纹理和光谱特征参数进行 LNC 估算。结果表明,融合模型在不同生育期的估算精度显著优于单一传感器模型,R2 提高了0. 03 ~ 0. 16,RMSE 降低了0. 02 ~ 0. 10 个百分点,NRMSE 减少了0. 61 ~ 3. 90 个百分点,其中SVR模型表现最佳。研究表明,多源遥感信息融合能够有效提升LNC估算精度,突破了单一传感器的局限,为精准农业提供了新的技术方法和理论支持。

Abstract:

Real-time monitoring of nitrogen status is essential for improving fertilizer use efficiency and crop yield and quality. Nitrogen monitoring plays a key role in precision agriculture; however, traditional soil and leaf sampling methods cannot meet the requirements for large-scale, real-time monitoring. Multi- source data from RGB and multi-spectral sensors were integrated to develop a model for estimating leaf nitrogen content ( LNC) in winter wheat, with the aim of improving the accuracy and reliability of nitrogen monitoring. Multi-source features were first extracted from UAV-based RGB and multi-spectral imagery acquired at key growth stages. Key variables were then identified by using a two-stage feature selection strategy that combined Pearson correlation analysis and variable importance in projection (VIP). Color space, texture, and spectral features were subsequently integrated to construct LNC estimation models. The results showed that the fused models consistently outperformed single-sensor models across different growth stages. Among the tested methods, the support vector regression (SVR) model achieved the best performance, with R2 increased by 0. 03 ~ 0. 16, RMSE decreased by 0. 02 ~ 0. 10 percentage points, and NRMSE reduced by 0. 61 ~ 3. 90 percentage points. These findings demonstrated that multi-source remote sensing data fusion can effectively improve LNC estimation accuracy and model robustness. The research result showed that multi-source remote sensing information fusion can effectively improve the accuracy of LNC estimation, overcome the limitations of a single sensor, and provide technical approaches and theoretical support for precision agriculture.