多光谱与热红外数据融合在冬小麦产量估测中的应用

为了解多光谱与热红外数据融合对冬小麦产量估测精度的影响，以30个黄淮麦区冬小麦品种为材料，利用三种灌溉处理（处理1、处理2和处理3灌水量分别为240、190和145 mm）下冬小麦拔节期、挑旗期、抽穗期与灌浆期的无人机多光谱和热红外动态数据，构造了多个光谱指数，以支持向量机构建冬小麦产量估测模型，并验证其精度。结果表明，植被指数与籽粒产量的相关性受溉水量影响，处理1下植被指数与籽粒产量均呈正相关，处理2下植被指数除土壤调整植被指数（SAVI）和转化叶绿素吸收反射指数（TCARI）外均与籽粒产量呈正相关，处理3下植被指数与籽粒产量均呈负相关。通过多光谱和热红外数据融合构建的冬小麦产量估测模型的预测精度比仅使用多光谱数据构建的模型提高8%。不同灌溉条件下，通过多光谱与热红外数据融合构建的模型的预测精度存在差异，在处理1、处理2和处理3下拔节期、挑旗期、抽穗期和灌浆期验证决定系数（R）最高值分别为0.63、0.68和0.56，均方根误差（RMSE）最低值分别为0.60、0.24和0.41 t·hm^-2，且在三种灌溉条件下灌浆期预测效果均最佳。因此，利用无人机光谱对小麦品种产量估测时应将多光谱与热红外数据融合，用支持向量机（SVM）算法构建产量估测模型，且模型在灌浆期具有较高预测  精度。

Application of Multispectral and Thermal Infrared Data Fusion in Estimation of Winter Wheat Yield

The construction of wheat yield prediction model based on multispectral and thermal infrared data fusion has important application value to improve the accuracy of grain yield estimation. In this study,30 winter wheat varieties from the Huanghuai wheat region were planted under three irrigation treatments(240 mm water amount under treatment 1 190 mm under treatment 2,and 145 mm under treatment 3). Infrared dynamic phenotypic data of wheat development were used to construct a yield estimation model,and the estimation accuracy was verified. The results showed that the vegetation index was affected by different irrigation treatments. Vegetation index of treatment 1 was positively correlated with grain yield; vegetation index of treatment 2 was positively related to grain yield except for soil adjusted vegetation index(SAVI) and transformed chlorophyll absorption reflectance index(TCARI). The vegetation index of treatment 3 was negatively correlated with the grain yield. The accuracy of constructing a winter wheat yield prediction model based on the fusion of multi-spectral and thermal infrared data is 8% higher than that of the prediction model using only single multi-spectral data. Under different irrigation conditions,there were differences in the accuracy of multi-spectral and thermal infrared data fusion to estimate yield. Under the three irrigation conditions,the determination coefficients(R) at the jointing stage,the flag picking stage,the heading stage and the filling stage were increased to 0.63,0.68 and 0.56,respectively,with the root mean square error(RMSE) reduced to 0.60 t·hm^-2,0.24 t·hm^-2 and 0.41 t·hm^-2,respectively.Under the three irrigation conditions,the prediction effect at grain filling stage is the best. Therefore,multi-spectral and thermal infrared data should be combined when using UAV spectroscopy to estimate the yield of wheat varieties. The yield estimation model constructed by the SVM algorithm has higher accuracy at the grain filling stage.