基于冠层RGB图像的冬小麦氮素营养指标监测

Estimation of wheat nitrogen nutrition indices in winter wheat based on canopy RGB images

[目的]本文旨在探索基于冬小麦冠层RGB图像的氮素营养指标实时监测方法,为实现简便、准确的冬小麦氮素营养诊断和推荐施肥奠定基础。[方法]基于3年次的冬小麦大田氮肥梯度试验,采用数码相机在返青期和拔节期垂直拍摄冠层RGB图像。分析图像特征参数绿红通道比值(G/R)、绿红通道差值(GMR)、红光标准化值(NRI)、绿光标准化值(NGI)、色相(H)和冠层覆盖度(CC)与植株氮素生理指标间的关系,筛选氮素营养监测指标的最优图像特征参数,构建氮素营养指标估算模型。[结果]CC与冬小麦地上部生物量、氮积累量和叶面积指数(LAI)三者间的相关系数最高,分别为0.87、0.85和0.84(P<0.01);其他特征参数与三者间的相关系数相对较低,其中H为0.81、0.77和0.79,NRI为-0.80、-0.77和-0.77,G/R为0.73、0.63和0.76,GMR为0.66、0.67和0.63。采用CC作为冬小麦氮素营养指标估算模型的输入参数,并分别使用异速生长函数和指数函数建立地上部生物量、氮积累量和LAI估算模型,异速生长函数这3个指标的估算模型R2分别为0.82、0.76和0.82(P<0.01),指数函数的R2分别为0.80、0.74和0.85(P<0.01)。利用独立试验数据对模型进行验证,异速生长函数模型预测值和观测值间的R2平均为0.89(P<0.01),地上部生物量、氮积累量和LAI预测值的均方根误差(RMSE)分别为31.09 g·m^-2、1.37 g·m^-2和0.16;指数函数模型预测值和观测值间的R2平均也为0.89(P<0.01),地上部生物量、氮积累量和LAI预测值的RMSE分别为28.95 g·m^-2、1.34 g·m^-2和0.17。[结论]异速生长函数和指数函数模型在利用CC对冬小麦氮素营养指标进行估算时均具有较好的预测性。基于RGB图像的监测方法操作简单、准确度高,可实时获取监测结果,具有较高的推广应用价值。

[Objectives]This study is expected to explore the real-time estimation of nitrogen(N)nutrition indices based on winter wheat canopy RGB images,which will lay a foundation for the simple and accurate diagnosis of N status and the recommendation of fertilization.[Methods]Based on 3 years/varieties N gradient trials of winter wheat in the fields,canopy RGB images were taken by a digital camera at returning green stage and jointing stage.The relationships were analyzed between image feature parameters(the ratio of green and red channel,G/R;the difference of green and red channel,GMR;normalized redness intensity,NRI;normalized greenness intensity,NGI;hue,H;canopy coverage,CC)and crop N-related indices(shoot dry matter,shoot N accumulation and leaf area index).The N nutrition estimation models were established based on the optimal image feature parameters.[Results]The image feature para-meter CC(canopy coverage)had the highest correlation coefficients with shoot dry matter,shoot N accumulation and LAI in winter wheat,which reached 0.87,0.85 and 0.84(P<0.01),respectively,compared with image color feature parameters H(0.81,0.77 and 0.79,respectively),NRI(-0.80,-0.77 and-0.77,respectively),G/R(0.73,0.63 and 0.76,respectively)and GMR(0.66,0.67 and 0.63,respectively)(P<0.01).Since the exponential relationships were found between CC and crop N-related indices,both allometric function and exponential function were chosen to establish the shoot dry matter,shoot N accumulation and LAI estimation models with CC.The determination coefficient(R2)in allometric function were 0.82,0.76 and 0.82(P<0.01)for shoot dry matter,shoot N accumulation and LAI,respectively,and R2 in exponential function were 0.80,0.74 and 0.85(P<0.01),respectively.Calibration was conducted on the two models with an independent dataset.The root mean square errors(RMSE)of shoot dry matter,shoot N accumulation and LAI in the allometric model were 31.09 g·m^-2,1.37 g·m^-2 and 0.16,respectively,with an average R2 of 0.89(P<0.01)between observed values and predicted values.The RMSE of shoot dry matter,shoot N accumulation and LAI in the exponential model were 28.95 g·m^-2,1.34 g·m^-2 and 0.17,respectively,with the same average R2 of 0.89(P<0.01)between observed values and predicted values.[Conclusions]These results indicate that allometric model and exponential model both give good predictions on N-related indices in winter wheat by using CC.The method of using CC in canopy RGB images for N-related indices estimation in winter wheat is simply operating with high accuracy and can obtain results in real-time,which has the potential of broadly application.