基于无人机高光谱的马铃薯冠层叶片全氮含量反演  

Inversion of leaf nitrogen content in potato canopy based on unmanned aerial vehicle hyperspectral images

作　　者：郭发旭 冯全 杨森 杨婉霞 GUO Faxu;FENG Quan;YANG Sen;YANG Wanxia(Mechanical and Electrical Engineering College,Gansu Agriculture University,Lanzhou 730070,China)

机构地区：[1]甘肃农业大学机电工程学院,甘肃兰州730070

出　　处：《浙江农业学报》2023年第8期1904-1914,共11页Acta Agriculturae Zhejiangensis

基　　金：国家自然科学基金(32160421);甘肃省教育厅产业支撑项目(2021CYZC-57);甘肃省优秀研究生“创新之星”项目(2022CXZXS-012);甘肃省高等学校青年博士基金(2021QB-033);甘肃省青年科技基金(20JR10RA544)。

摘　　要：为实现大田马铃薯冠层叶片全氮含量(LNC)的快速反演,利用低空无人机平台搭载成像光谱仪获取马铃薯冠层光谱数据,在综合比较原始反射率(R)、倒数变换反射率(1/R)、一阶微分变换反射率[D(R)]、二阶微分变换反射率[D(2 R)]、倒数之对数变换反射率[lg(1/R)]的基础上,选择[D(2 R)]用于后续试验。分别使用相关性分析(CA)、竞争性自适应重加权(CARS)、无信息变量消除(UVE)3种算法筛选特征光谱波段,使用偏最小二乘回归(PLSR)、支持向量机(SVM)构建马铃薯冠层LNC估测模型。结果表明:CA、CARS、UVE算法分别筛选出26、12、19个特征波段。在构建的PLSR模型中,用UVE筛选的特征波段建立的预测模型[UVE-D(2 R)-PLSR]效果最好,在验证集上的决定系数(R^(2))和均方根误差(RMSE)分别为0.8068和0.1932;在构建的SVM模型中,用CARS筛选的特征波段建立的预测模型[CARS-D(2 R)-SVM]效果最好,在验证集上的R 2和RMSE分别为0.8316和0.1830。两模型对比,CARS-D(2 R)-SVM模型的效果更好。采用CARS-D(2 R)-SVM模型逐点估算马铃薯冠层LNC,绘制反演图,可使种植者直观掌握大田马铃薯生长情况,为马铃薯大田的精细化管理提供数据支持。To realize the rapid inversion of leaf nitrogen content(LNC)in the canopy of field potatoes,the spectral data of potato canopy leaves were obtained by an imaging spectrometer of a low-altitude unmanned aerial vehicle(UAV)platform.Based on the comprehensive comparison of original reflectance(R),reciprocal transformation reflectance(1/R),first-order differential transformation reflectance[D(R)],second-order differential transformation reflectance[D(2 R)],and logarithm of reciprocal transformation reflectance[lg(1/R)],[D(2 R)]was selected for the subsequent experiment.Correlation analysis(CA),competitive adaptive reweighed sampling(CARS)and uninformative variables elimination(UVE)algorithms were introduced to screen the characteristic spectral bands,and partial least squares regression(PLSR)and support vector machine(SVM)algorithms were used to construct the LNC estimation model.It was shown that 26,12 and 19 characteristic bands were screened out by CA,CARS and UVE algorithms,respectively.Among all the established PLSR models,the one based on characteristic bands sreend out by UVE[UVE-D(2 R)-PLSR for short]had the best performace,as its determinatino coefficient(R 2)and root mean square error(RMSE)on the validatin set were 0.8068 and 0.1932,respectively.Among all the established SVM models,the one based on characteristic bands screened out by CARS[CARS-D(2 R)-SVM for short]had the best performance,as its R^(2) and RMSE on the validation set were 0.8316 and 0.1830,respectively.Compared with UVE-D(2 R)-PLSR,CARS-D(2 R)-SVM showed better modeling effect.The constructed CARS-D(2 R)-SVM model was used to estimate LNC based on the spectral image of potato canopy,and the inverse diagram of LNC was plotted,which could help the growers intuitively grasp the potato growth in the field and provide data support for the potato field management.

关 键 词：无人机 高光谱 马铃薯 支持向量机 反演 

分 类 号：S127[农业科学—农业基础科学]