基于无人机数码影像的水稻叶面积指数监测

Monitoring Rice Leaf Area Index Based on Unmanned Aerial Vehicle(UAV) Digital Images

【目的】为探究无人机数码影像监测水稻叶面积指数(Leaf area index,LAI)的可行性,明确利用无人机数码影像监测水稻LAI的最佳时期,构建基于无人机数码影像的水稻LAI监测模型。【方法】本研究基于不同品种和施氮量的水稻田间试验,于分蘖期、拔节期、孕穗期、抽穗期和灌浆期测定水稻LAI,同步使用无人机搭载数码相机获取水稻无人机数码影像并提取颜色指数及纹理特征,分析其在不同生育时期与水稻LAI之间的相关性,构建定量监测模型,并用独立试验数据对所建模型进行检验。【结果】无人机数码影像中颜色指数及纹理特征与水稻LAI之间的相关性在生育前期(分蘖期+拔节期)最高,高于所有单生育期、生育后期(孕穗期+抽穗期+灌浆期)和全生育期,可确定为监测的最佳时期;在颜色指数和纹理特征当中,纹理特征方差(Variance,VAR)在监测水稻生育前期LAI时表现最优,可构建监测模型LAI=1.1656×exp^((0.0174×VAR))实现监测,模型构建时的决定系数(Determination coefficient,R^(2))为0.7980,模型检验时的相对均方根误差(Relative root mean square error,RRMSE)和偏差(bias,θ)分别为0.1658和0.1306。【结论】与人工测量LAI相比,基于无人机数码影像的水稻LAI监测方法可提高作业效率,降低成本,在水稻长势快速准确监测和丰产高效栽培中具有应用价值。

【Obiective】Leaf area index (LAI) is a crucial variable for assessing rice growth,and unmanned aerial vehicle(UAV) digital images can serve as an efficient way to real-time,no-destructive monitoring of crop growth parameters.However,it remains unclear which parameter in digital images can be used to estimate rice LAI.In addition,the optimal growth stage for monitoring is also unknown.【Method】In this study,the UAV digital images were initially collected from two field experiments encompassing variations over two years with four cultivars at four nitrogen application levels.Then,the relationship between UAV digital image parameters (nine color indices and eight texture features) and rice LAI at different growth stages (tillering stage,jointing stage,booting stage,heading stage and filling stage) were analyzed.【Result】The results suggested that the early growth stages,including both tillering stage and jointing stage,were suitable for rice LAI monitoring through UAV digital images,and the texture feature variance (VAR) exhibits greatest accuracy in model calibration with a determination coefficient (R^(2)) of 0.7980.In the validation based on independent experiment,this texture feature also performs well with relative root mean square error (RRMSE) of 0.1658 and bias(θ)of 0.1306.【Conclusion】Taking the accuracy and convenience in application into consideration,we found that the texture feature VAR could be used to monitor rice LAI in early growth stage with estimation models of LAI =1.1656×exp^((0.0174×VAR)).