多类型参数结合的冬小麦拔节期SPAD值估算

Combining Multiple Type Parameters for Estimating Winter Wheat SPAD Values at Jointing Stage

摘要:

冬小麦拔节期叶绿素状况估算对于冬小麦营养诊断非常重要。本文利用无人机遥感平台获取冬小麦拔节期长势遥感信息，提取多光谱植被指数、RGB影像纹理特征及覆盖度信息，基于多元线性回归（Multivariate linear regression，MLR）、随机森林回归（Random forest regression，RFR）构建了冬小麦SPAD值估算模型。分析多光谱植被指数、纹理特征和覆盖度信息，以及相互结合对于冬小麦SPAD值估算的影响。结果表明：多光谱植被指数、纹理特征、覆盖度的结合（2种类型或3种类型参数结合）可以用于冬小麦拔节期SPAD值的估算，而且相较于单类型参数或两类型参数结合，更多类型参数结合提高了冬小麦拔节期SPAD值的估算精度。而基于相同参数利用随机森林构建的冬小麦拔节期SPAD值估算模型精度均高于多元线性回归构建的模型精度。其中，基于3种类型参数构建的冬小麦SPAD值估算模型精度最高，R2为0.78，RMSE为2.08。各类型参数对冬小麦拔节期SPAD值估算精度的影响由大到小依次为多光谱植被指数、纹理特征、覆盖度。其中，多光谱植被指数构建的模型精度与纹理特征构建的模型精度相近（R2和RMSE分别为0.71、2.36及0.70、2.45）。覆盖度虽然对于SPAD值的估算精度提升最小，但结合其他特征可提高冬小麦SPAD值估算精度（对于RFR模型，R2提高0.02~0.03）。多光谱植被指数、纹理特征、覆盖度的结合提高了模型估算精度，为冬小麦拔节期SPAD值快速估算提供了技术参考。

Abstract:

Estimation of chlorophyll status at jointing stage of winter wheat is important for nutritional diagnosis of winter wheat. The UAV remote sensing platform was used to obtain the remote sensing information of winter wheat growth at the jointing stage, the multispectral vegetation indexes, RGB image texture features and coverage information were extracted, the models for estimating the winter wheat SPAD values were constructed based on the multivariate linear regression (MLR) and random forest regression (RFR). Then the effects of multispectral vegetation indexes, textural features, cover information, and combining them with each other on the estimation of winter wheat SPAD values were analyzed. The results showed that the combination of multispectral vegetation indexes, texture features, canopy coverage (combination of two or three types of parameters) can be used for the estimation of winter wheat SPAD values at the jointing stage, and the combination of more types of parameters improved the estimation accuracy of winter wheat SPAD values compared with the combination of single-type parameters or two types of parameters. And the accuracy of the winter wheat SPAD estimation model constructed based on the same parameters by using RFR was higher than that of the model constructed by MLR. Among them, the model constructed based on the three types of parameters had the highest estimation accuracy of winter wheat SPAD values, with R2 of 0.78 and RMSE of 2.08. Moreover, the effects of each type of parameters on the models accuracy improvement in descending order were multispectral vegetation indexes, texture features, and canopy coverage. Among them, the accuracy of the model constructed by multispectral vegetation indexes was similar to that of the model constructed by texture features. Canopy coverage had the smallest improvement in the estimation accuracy of SPAD values, but combining other features could improve the estimation accuracy of winter wheat SPAD values (R2 was increased by 0.02~0.03 for the RFR models). The combination of multispectral vegetation indexes, texture features and canopy coverage improved the accuracy of the models, providing a fast technical reference solution for winter wheat SPAD values estimation.