基于无人机影像的小麦株高与LAI预测研究

为快速、准确地估测不同生育时期小麦品种（系）株高与叶面积指数（LAI）表型性状，基于各生育时期小麦品种（系）数字正射影像（digital orthophoto map，DOM）和数字表面模型（digital surface model，DSM），分别构建不同生育时期株高估测模型和光谱指数LAI估测模型。借助一元线性回归、多元逐步回归（SMLR）和偏最小二乘回归（PLSR）分析方法，并采用决定系数（r）、均方根误差（RMSE）和归一化均方根误差（nRMSE）综合性评价指标，筛选出小麦不同生育时期最优的株高和LAI估测模型。结果表明，（1）全生育期株高估测效果最好，模型预测值与实测值高度拟合（r、RMSE、nRMSE分别为0.87、5.90 cm、9.29%）；在各生育时期中，灌浆期模型预测精度较好，成熟期预测精度最差，r分别为0.79和0.69。（2）所选的18种光谱指数与LAI相关性均较好，其中BGRI、RGBVI、NRI和NGRDI的相关系数达到极显著水平，且各时期三种回归估测模型均表现出较高的稳定性和拟合效果，其中SMLR回归模型对各生育时期LAI预测精度最好，其拔节期、孕穗期、扬花期、灌浆期和成熟期的预测集r分别为0.68、0.57、0.61、0.68和0.53。这说明，基于无人机获取的不同生育时期小麦DSM影像提取株高，并运用18种光谱指数构建LAI估测模型方法是可行的。

Prediction of Wheat Plant Height and Leaf Area Index Based on UAV Image

In order to quickly and accurately estimate plant height and leaf area index(LAI) phenotypic characters of wheat varieties(lines), the LAI estimation model and spectral index estimation model of plant height at different growth periods were constructed based on the digital orthophoto map(DOM) and digital surface model(DSM). With the help of single linear regression, multiple stepwise regression(SMLR) and partial least squares regression(PLSR) analyses, the comprehensive evaluation indices of determination coefficient(r), root mean square error(RMSE) and normalized root mean square error(nRMSE), the best model for plant height and LAI estimation in different growth periods was selected. The results showed that the model of plant height estimation in the whole growth period had the best effect, and its predicted value of plant height was highly fitted with the measured value(r, RMSE and nRMSE were 0.87, 5.90 cm and 9.29%, respectively); in each growth period, the prediction accuracy of the model in the filling period was better, while that at maturity was the worst; r was 0.79 and 0.69, respectively. The correlation coefficients of BGRI, RGBVI, NRI and NGRDI were significant, and the three regression estimation models in each period show high stability and fitting effect, among which SMLR regression model had the best prediction accuracy for LAI in each growth period, and its prediction of jointing, booting, flowering, filling and maturity periods is the best; r was 0.68, 0.57, 0.61, 0.68 and 0.53, respectively. This shows that it is feasible to extract plant height from DSM images of wheat at different growth stages obtained by UAV and build LAI estimation model by using 18 spectral indices.