码上扫一扫！ |
| 基于无人机多光谱影像的油茶冠层氮磷钾含量估算模型 |
|
徐丹丹1,2, 段丹丹3,4, 陈龙跃3,4, 赵春江3,4
|
|
1.江苏大学 农业工程学院，江苏 镇江 212013;2.南京林业大学 生态与环境学院， 江苏 南京 210037;3.北京市农林科学院 信息技术研究中心，北京 100097;4.国家农业信息化工程技术研究中心，北京 100097
|
|
| 摘要: |
| 【 目的】基于无人机多光谱影像构建油茶冠层氮、磷、钾含量诊断模型，为油茶施肥管理提供参考。 【方法】以广东省河源市的美林湖油茶样地为研究对象，基于成熟油茶叶片的氮、磷、钾含量和无人机多光谱影像（Yunsense MS600V2多光谱传感器），选取39个植被指数，采用竞争适应性重加权采样（CARS）选择特征植被指数，构建氮、磷、钾含量估算的极限学习机（ELM）、岭回归（RR）和偏最小二乘法（PLSR）模型，并采用留一交叉验证法，以 均方根误差（RMSE）、相对均方根误差（RRMSE）、决定系数（R2）以及估算结果的空间分布和数值分布评价模型精 度。【 结果】CARS选择结果显示，油茶冠层N含量估算的特征植被指数有9个，分别为EXB、TGI、BNDVI、ENDVI、DATT750、NDREI、RECI、SCCCI750、TVI；P含量估算的特征植被指数有7个，分别为EXB、TGI、ENDVI、GNDVI、NDVI、MCARI750、TVI；K含量估算的特征植被指数有6个，分别为IKAW、TGI、VARI、NDVI、NDRE、RECI750；油茶冠层N含量估算模型RR和PLSR的R2（0.528和0.669）虽略低于ELM模型（0.679），但其RRMSE分别为5.08%和5.23%，低于ELM的5 53%，且以RR模型估算结果的平均值、标准差和范围更接近于野外实测数据；油茶冠层P和K含量估算模型中，RR模型的估算结果平均值、标准差和范围更接近野外实测数据，估算效果优于PLSR和ELM模型。【 结论】基于无人机多光谱影像，并结合CARS选择特征植被指数和RR模型，可实现油茶冠层氮磷钾含量低成 本、高频率、高效率的无损快速诊断，为油茶施肥管理和提高产量提供了有效的技术支持。 |
| 关键词: 油茶 冠层氮磷钾含量 无人机 多光谱影像 营养诊断 |
| DOI：10. 13207/j. jnwafu. 2026. 09. 003 |
| 分类号: |
| 基金项目:岭南现代农业科学与技术广东省实验室河源分中心项目（DT20220001）；国家重点研发计划项目（2022YFD2202102）；广东 省科技专项资金项目（210909114530725）；广东省科技计划项目（2023B0208010002） |
|
| Estimation model of the canopy N，P，K content for Camellia oleifera based on UAV multispectral images |
|
XU Dandan1,2, DUAN Dandan3,4, CHEN Longyue3,4, ZHAO Chunjiang3,4
|
|
1.School of Agricultural Engineering，Jiangsu University，Zhenjiang，Jiangsu 212013，China;2.College of Ecology and Environment，Nanjing Forestry University，Nanjing，Jiangsu 210037;3.Information Technology Research Center，Beijing Academy of Agriculture and Forestry Sciences，Beijing 100097，China;4.National Engineering Research Center for Information Technology in Agriculture，Beijing 100097，China
|
| Abstract: |
| 【 Objective】Based on UAV multispectral images，this study aims to build the diagnostic models for estimating canopy nitrogen（N），phosphorus（P），and potassium（K） content in Camellia oleifera，which provides references for fertilization management.【Method】This study was conducted in a C.oleifera plantation in Meilinhu，Heyuan city，Guangdong province. The characteristic vegetation indices were selected based on competitive adaptive reweighted sampling （CARS） method and 39 vegetation indices were calculated from UAV multispectral images（Yunsense MS600V2 sensor）.Extreme Learning Machine（ELM），Partial Least Square Regression Model（PLSR） and ridge regression（RR） models were built to estimate canopy N，P，K contents based on the field collected mature leaves and CARS selected vegetation indices. The leave-one-out cross validation method was used for accuracy assessment，with performance metrics including RMSE，RRMSE，and R2，supplemented by an analysis of the spatial and numerical distribution of the estimates.【Result】The CARS method selected 9（EXB，TGI，BNDVI，ENDVI，DATT750，NDREI，RECI，SCCCI750，TVI），7 （EXB，TGI，ENDVI，GNDVI，NDVI，MCARI750，TVI） and 6（ IKAW，TGI，VARI，NDVI，NDRE，RECI750） characteristic vegetation indices for estimating canopy N，P，K contents，respectively. For canopy N content estimation，the R2 values of the RR（0.528） and PLSR（0.669） models were slightly lower than that of the ELM model（0.679），but both models had lower RRMSE（RR：5.08%；PLSR：5.23%） than the ELM model （5.53%）.Furthermore，the mean，stand deviation，and range of the RR model’s estimates were closer to the field-measured data. For estimating canopy P and K contents，the RR model demonstrated superior performance over the ELM and PLSR models，as its estimates showed the closest agreement with the field-measured data in terms of mean，standard deviation，and range.【Conclusion】This study provides a non-destructive，low-cost，high-frequency，and high-efficiency method for rapid canopy N，P，K estimation of C.oleifera. By integrating UAV multispectral images with vegetation indices selected via CARS and the RR model，it provides a practical tool to support precision fertilization and yield enhancement. |
| Key words: Camellia oleifera canopy N,P,K content UAV multispectral images nutrition diagnosis |
