不同分辨率无人机多光谱影像的棉花叶面积指数估测研究

Estimation of leaf area index of cotton from unmanned aerial vehicle multispectral images with different resolutions

【目的】叶面积指数(leaf area index,LAI)是农业生产中用于作物生长诊断、生物量估算和产量估测的重要指标之一。利用无人机多光谱遥感数据快速反演LAI对棉花长势诊断和田间管理具有重要意义。【方法】以新疆阿拉尔垦区花铃期棉花为研究对象,以地面实测LAI数据及无人机多光谱影像为数据源,进行影像拼接,然后对高空间分辨率无人机光谱影像进行6种不同分辨率的空间重采样,分别提取植被指数和纹理特征,并用纹理特征构建纹理指数,以植被指数、纹理指数和二者融合分别为输入量,基于偏最小二乘回归(partial least squares regression,PLSR)、支持向量机(support vector machine,SVM)和随机森林(random forest,RF)算法构建棉花LAI估测模型,分别比较不同分辨率下3种输入特征量与3种算法构建的模型估测精度。【结果】(1)随着多光谱影像分辨率的降低,植被指数和纹理指数与LAI的相关性均呈现先上升后下降的趋势,影像分辨率为1.0 m时,二者与LAI的相关性最高。不同模型的估测精度也随影像分辨率的降低呈先上升后下降的趋势,1.0 m分辨率下估测效果最好。(2)1.0 m分辨率多光谱影像下,RF算法模型估测效果最佳,其次为SVM算法模型,PLSR算法模型估测效果最差。(3)3种输入特征量对棉花LAI估测效果的优劣顺序依次为:植被指数与纹理指数融合、植被指数、纹理指数。【结论】利用1.0 m空间分辨率的无人机多光谱遥感影像提取的植被指数与纹理指数构建的RF算法模型,可以实现对棉花花铃期LAI的高精度估测。

[Objective]Leaf area index(LAI)is one of the important indexes for crop growth diagnosis,biomass estimation and yield prediction in agriculture.Rapid inversion of LAI using unmanned aerial vehicle(UAV)multispectral remote sensing is important for cotton growth monitoring and field management.[Methods]Cotton at flowering and boll-setting stage in Aral reclamation area in Xinjiang was taken as the research object.The ground measured cotton LAI data and UAV multispectral images were used as data sources.After the image stitching was completed,spatial resampling was performed to obtain six different resolutions of multispectral images,vegetation index and texture features were extracted,and the texture index was constructed with texture features.Using the vegetation index,texture index and the combination of the two indices as input quantities,the cotton LAI prediction model based on the partial least squares regression(PLSR),support vector machine(SVM)and random forest algorithm(RF)were constructed and the prediction performance of the three input features and the models were compared under different resolutions,respectively.[Results](1)The correlation between the two indices and LAI tended to increase and then decrease when the resolution of the multispectral image decreased,and the correlation between the two indices and LAI reached maximum when the image resolution was under the 1.0 m.The estimation performance of the three models were first increased and then decreased with the reduction of image resolution,and the estimation performance was the best under the 1.0 m resolution.(2)When the multispectral image was under the 1.0 m resolution,the RF algorithm model has the best estimation results,followed by the SVM algorithm model,and the PLSR algorithm model has the worst results.(3)The order of the accuracy of three input feature quantities were the follow:vegetation index and texture index,vegetation index,and texture index.[Conclusion]The RF algorithm model constructed by vegetation index and texture index extracted from UAV remote sensing image with 1.0 m spatial resolution had the highest accuracy in estimating cotton LAI at flowering and boll-setting stage.