基于数码相机的玉米冠层SPAD遥感估算

【目的】叶绿素是植物光合作用中重要的色素。利用作物光谱信息对叶绿素含量进行反演,为作物的实时监测和生长状态诊断提供重要依据。【方法】以大田环境下不同氮肥水平(0,50%和100%)的开花期玉米为研究对象,利用轻小型无人机搭载数码相机,获取试验区RGB影像。使用土壤调整植被指数(soil adjusted vegetation index,SAVIgreen)对图像进行分割,基于分割前后的影像分别提取15种常见的可见光植被指数,综合分析指数与玉米冠层叶绿素相对含量SPAD值的相关关系。采用单变量回归模型、多元逐步回归模型和随机森林(random forest,RF)回归算法构建玉米SPAD值的遥感估算模型,通过模型精度评价指标决定系数(coefficient of determination,R2)、均方根误差(root mean square error,RMSE)、平均相对误差(mean relative error,MRE)和显著性检验水平(P0.01),确定最佳指标和最优模型。【结果】基于分割前后的数码影像提取的VIplot和VIplant植被指数与玉米冠层SPAD值之间具有显著的相关关系,其中VIplant中的红光标准化值(NRI)、归一化叶绿素比值植被指数(NPCI)、蓝红比值指数(BRRI)、差值植被指数(DVI)与SPAD值的相关性在0.77以上;以相关性高于0.77的VIplant指数NRI、NPCI、BRRI、DVI构建的线性、指数、对数、二次多项式、幂函数的单变量回归模型中,NRI指数构建的二次多项式模型效果最好,决定系数R2为0.7976,RMSE为4.31,MRE为5.91%。在VIplant指数NRI、NPCI、BRRI、DVI参与建立的多变量SPAD反演模型中,使用随机森林方法的模型精度最高,决定系数R2为0.8682,RMSE为3.92,MRE为4.98%,而多元逐步回归模型的精度高于任意单变量回归模型,决定系数R2为0.819,RMSE为4,MRE为5.67%;对数码影像结合各模型制作的SPAD分布图进行精度分析,使用随机森林回归模型对SPAD的估测值与实测值最为接近,具有最佳的预测效果,R2为0.8247,RMSE为4.3,MRE为5.36%,可以作为玉米冠层叶绿素信息监测的主要方法。【结论】本研究证明将数码相机影像提取的可见光植被指数应用于玉米叶绿素相对含量的估测是可行的,这也为无人机遥感系统在农业方面的应用增添了新的手段和经验。

Remote Sensing Estimation of Canopy SPAD Value for Maize Based on Digital Camera

【Objective】Chlorophyll is an important pigment in plant photosynthesis. The objective of this study is to investigate the inversion of chlorophyll content using crop spectrum information, so as to provide an important basis for real-time monitoring and diagnosis of crop growth.【Method】Based on the field environment under different nitrogen fertilizer application levels (0, 50% and 100%) of maize, the light and small UAV equipped with consumer level digital camera was used to obtain the RGB image of the test area, and then the soil adjusted vegetation index was used for image segmentation. 15 common visible vegetation indexes were extracted based on images before and after segmentation. Then the correlation between vegetation index and SPAD values were analyzed, besides single variable regression model, multiple regression model and random forest regression model based on visible vegetation indexes were established to estimate the SPAD values. And then, the indicators of accuracy evaluation, coefficient of determination, root mean square error, mean relative error and P<0.01 were used to select the best indicators and the optimal model.【Result】There was a significant correlation between VI_plot and VI_plantvegetation indexes and the SPAD value of maize canopy, for example, the correlation coefficient between normalized redness intensity (NRI), normalized pigment chlorophyll ratio index (NPCI), blue red ratio index (BRRI) and SPAD value of VI_plant was above 0.77. The univariate regression models were built, which took NRI,NPCI, BRRI and DVI as the independent variables and the measured SPAD as dependent variable, including linear, exponential, logarithmic, two degree polynomial and power function models, and among those models, the two polynomial model constructed by the NRI index was the best one with the decision coefficient R² of 0.7976, the RMSE of 4.31, and the MRE of 5.91%; the precision of the model using the random forest regression algorithm was the highest, in which the determining coefficient was 0.8682, the RMSE was 3.92, and the MRE was 4.98%; the multiple regression model had higher accuracy than any single variable regression model, in which the decision coefficient R² was 0.819, RMSE was 4, and MRE was 5.67%. The six inversion models of SPAD were used to make the distribution map of corn canopy SPAD value, and then the map using random forest regression model had the best result which was the closest to real SPAD distribution with R² of 0.8247，RMSE of 4.3，MRE of 5.36％, therefore which could be used as a main method of corn canopy chlorophyll monitoring information.【Conclusion】The results showed that the application of UAV digital imagery in retrieving SPAD of corn was feasible, which also added new means and experience to the application of UAV remote sensing system in agriculture.