基于无人机多光谱的不同施硒水平荞麦冠层叶绿素估测模型研究

Study on Canopy Chlorophyll Estimation Model of Buckwheat with Different Selenium Levels Based on UAV Multispectrum

为验证采用低空无人机(UAV)搭载多光谱相机进行荞麦冠层叶绿素含量估测的可能性,同时探索硒元素(Se)对荞麦冠层叶绿素及光谱特征的影响规律,以晋荞6号甜荞和晋荞9号苦荞为研究对象,在山西省晋中市太谷区开展了试验研究。通过无人机搭载多光谱相机采集了不同施硒水平下荞麦冠层的多光谱图像,并同步在田间测得叶绿素相对含量,首先分析了不同施硒水平下2种荞麦冠层叶绿素含量的变化规律,其次通过提取多光谱图像的光谱信息获得荞麦5个波段下的光谱信息,在此基础上分析了荞麦在盛花期与灌浆期的光谱特征规律,运用5个波段下的反射率构建了11种植被指数,将16个光谱变量与实测冠层叶片叶绿素含量进行了皮尔逊相关性分析,采用偏最小二乘法回归(PLSR)、主成分回归(PCR)、支持向量机回归(SVR)、反向传播神经网络(BPNN)构建了多光谱波段反射率-植被指数的荞麦叶绿素含量遥感监测模型,并通过精度检验确定最优估算模型。结果表明,适量喷施硒肥可增加荞麦叶绿素含量,过量施加会抑制叶绿素含量。5个波段中,蓝、红、红边、近红外均表现出较强的相关性,其中近红外波段相关性高且较为稳定;植被指数方面,标准化降水指数(SPI)、绿色叶绿素指数(GCI)、绿色归一化差异植被指数(GNDVI)、归一化绿光指数(NGI)、转换优化土壤调节植被指数(TOSAVI)、转换叶绿素吸收比指数(TCARI)、三角植被指数(TVI)这7种植被指数的|r|为0.50~0.91,存在较好的相关性。在盛花期,运用BPNN的预测效果最好,预测集相关系数R2P达0.97,预测集均方根误差RMSE为0.95;在灌浆期,运用SVR的预测效果优于其他模型,预测集R2P为0.96,RMSE为0.45;在开花期—灌浆期,PLSR表现最好,预测集R2P为0.98,RMSE为0.28。而就模型而言,SVR相较于其他模型表现出更高的稳定性和准确性,预测集R2P在0.94~0.96,RMSE在0.45~0.82,RPD均大于3.00。表明无人机低空遥感可实现田间荞麦冠层叶绿素含量的快速监测,为无人机低空预测荞麦叶绿素含量的模型算法优化提供了参考。

In order to verify the possibility of estimating the chlorophyll content of buckwheat canopy by using low⁃altitude unmanned aerial vehicle(UAV)with multi⁃spectral camera,and to explore the effect of selenium(Se)on the chlorophyll and spectral characteristics of buckwheat canopy,Jinsage No.6 sweet buckwheat and Jinsage No.9 bitter buckwheat were selected as the research objects,and a experimental study was conducted in Taigu District,Jinzhong City,Shanxi Province.Buckwheat multispectral images were collected under different selenium levels at different stages by using UAV with multi⁃spectral camera,and the relative chlorophyll content(SPAD value)was synchronously measured in the field.Firstly,the SPAD values of buckwheat canopy under different selenium application levels were analyzed.Secondly,the reflectance of buckwheat canopy under five bands was obtained by extracting the spectral information of multi⁃spectral images.On this basis,the spectral characteristics of buckwheat at full bloom and filling stage were analysed,11 vegetation indices were constructed using the reflectance in five bands,and the absolute magnitude of the correlation coefficients between the 16 spectral variables and the measured SPAD values were obtained by Pearson correlation analysis.Partial least square regression(PLSR),principal component regression(PCR),support vector machine regression(SVR)and back propagation neural network(BPNN)were used to construct a buckwheat canopy SPAD monitoring model,and the optimal estimation model was determined by accuracy test.The results showed that moderate application of selenium fertilizer could increase SPAD value of buckwheat,while excessive application could inhibit SPAD value.The blue,red,red⁃edge and NIR bands showed strong correlation among the five bands,and the NIR band showed high and stable correlation.In terms of vegetation index,the correlation coefficients(|r|)of standardized precipitation index(SPI),green chlorophyll index(GCI),green normalized difference vegetation index(GNDVI),normalized green light index(NGI),transformed optimized soil⁃regulated vegetation index(TOSAVI),transformed chlorophyll absorption ratio index(TCARI),and triangular vegetation index(TVI)ranged from 0.50 to 0.91,which indicated a good correlation.At the full⁃bloom stage,BPNN had the best prediction effect,the correlation coefficient of prediction set(R2P)was 0.97,and the root mean square error(RMSE)of prediction set was 0.95.In the filling stage,the prediction effect of SVR was better than other models,R2P was 0.96,and RMSE was 0.45.At the full⁃bloom stage and filling stage,PLSR showed the best performance,R2P was 0.98,RMSE was 0.28.By comparing all the models,SVR showed higher stability and accuracy(R2P and RMSE ranges were 0.94—0.96 and 0.45—0.82,respectively,and RPD values were greater than 3.00).These results show that UAV with a multispectral camera can achieve rapid monitoring of buckwheat canopy chlorophyll content in the field,providing a reference for optimization of the model algorithm for low altitude prediction of chlorophyll content by UAV.