College of Resources and Environment, Shandong Agricultural University, Tai'an 271018, China.
National Engineering Research Center for Efficient Utilization of Soil and Fertilizer Resources, Tai'an 271018, China.
Sensors (Basel). 2022 May 4;22(9):3503. doi: 10.3390/s22093503.
As the major nutrient affecting crop growth, accurate assessing of nitrogen (N) is crucial to precise agricultural management. Although improvements based on ground and satellite data nitrogen in monitoring crops have been made, the application of these technologies is limited by expensive costs, covering small spatial scales and low spatiotemporal resolution. This study strived to explore an effective approach for inversing and mapping the distributions of the canopy nitrogen concentration (CNC) based on Unmanned Aerial Vehicle (UAV) hyperspectral image data in a typical apple orchard area of China. A Cubert UHD185 imaging spectrometer mounted on a UAV was used to obtain the hyperspectral images of the apple canopy. The range of the apple canopy was determined by the threshold method to eliminate the effect of the background spectrum from bare soil and shadow. We analyzed and screened out the spectral parameters sensitive to CNC, including vegetation indices (VIs), random two-band spectral indices, and red-edge parameters. The partial least squares regression (PLSR) and backpropagation neural network (BPNN) were constructed to inverse CNC based on a single spectral parameter or a combination of multiple spectral parameters. The results show that when the thresholds of normalized difference vegetation index (NDVI) and normalized difference canopy shadow index (NDCSI) were set to 0.65 and 0.45, respectively, the canopy's CNC range could be effectively identified and extracted, which was more refined than random forest classifier (RFC); the correlation between random two-band spectral indices and nitrogen concentration was stronger than that of other spectral parameters; and the BPNN model based on the combination of random two-band spectral indices and red-edge parameters was the optimal model for accurately retrieving CNC. Its modeling determination coefficient (R) and root mean square error (RMSE) were 0.77 and 0.16, respectively; and the validation R and residual predictive deviation (RPD) were 0.75 and 1.92. The findings of this study can provide a theoretical basis and technical support for the large-scale, rapid, and non-destructive monitoring of apple nutritional status.
作为影响作物生长的主要养分，准确评估氮素对于精准农业管理至关重要。尽管基于地面和卫星数据的氮素监测技术已经取得了一些改进，但这些技术的应用受到成本高昂、覆盖空间范围小以及时空分辨率低等因素的限制。本研究旨在探索一种基于无人机高光谱图像数据反演和绘制典型中国苹果园冠层氮浓度（CNC）分布的有效方法。在研究中，使用 Cubert UHD185 成像光谱仪搭载在无人机上获取苹果冠层的高光谱图像。采用阈值法确定苹果冠层范围，以消除裸土和阴影背景光谱的影响。分析并筛选出与 CNC 敏感的光谱参数，包括植被指数（VIs）、随机两波段光谱指数和红边参数。基于单个光谱参数或多个光谱参数的组合，构建偏最小二乘回归（PLSR）和反向传播神经网络（BPNN）反演 CNC。结果表明，当归一化差异植被指数（NDVI）和归一化差异冠层阴影指数（NDCSI）的阈值分别设置为 0.65 和 0.45 时，可以有效地识别和提取冠层的 CNC 范围，比随机森林分类器（RFC）更精细；随机两波段光谱指数与氮浓度的相关性强于其他光谱参数；基于随机两波段光谱指数和红边参数组合的 BPNN 模型是准确反演 CNC 的最优模型。其建模决定系数（R）和均方根误差（RMSE）分别为 0.77 和 0.16；验证 R 和剩余预测偏差（RPD）分别为 0.75 和 1.92。本研究结果可为大规模、快速、无损监测苹果营养状况提供理论依据和技术支持。
