基于无人机遥感的果园冠层氮素估算及空间分析  

Nitrogen estimation and spatial analysis of orchard canopy based on UAV remote sensing

作　　者：李达岁 阮思奇 胡青青 张金智[1,2] 张亚昊 佃袁勇 胡春根[1,2] 刘永忠[1,2] 雷宏伟 周靖靖[1] LI Dasui;RUAN Siqi;HU Qingqing;ZHANG Jinzhi;ZHANG Yahao;DIAN Yuanyong;HU Chungen;LIU Yongzhong;LEI Hongwei;ZHOU Jingjing(College of Horticulture&Forestry Science,Huazhong Agriculture University,Wuhan 430070,China;National Key Laboratory for Germplasm Innovation and Utilization of Horticultural Crops,Wuhan 430070,China;Forestry Protection Center,Landscape Bureau of Hongshan District,Wuhan City,Wuhan 430070,China;Jiangxi Lümeng Technology Holdings Co.,Ltd.,Ganzhou 341600,China)

机构地区：[1]华中农业大学园艺林学学院,武汉430070 [2]果蔬园艺作物种质创新与利用全国重点实验室,武汉430070 [3]武汉市洪山区园林局林业防护中心,武汉430070 [4]江西绿萌科技控股有限公司,赣州341600

出　　处：《华中农业大学学报》2023年第4期158-166,共9页Journal of Huazhong Agricultural University

基　　金：国家重点研发计划项目(2019YFD1000104);国家自然科学基金项目(31901963);国家柑橘产业技术体系(CARS-26)。

摘　　要：为快速准确地获取植株冠层氮素含量及空间分布特征,对大尺度的果园进行精准动态的管理,以宽行窄株小冠模式、宽行窄株篱壁模式和传统栽培模式3种栽培模式的120棵柑橘树为研究对象,通过测定冠层氮素含量并提取无人机遥感影像多光谱数据中的纹理指数和植被指数,运用随机森林算法(RF)建立基于植被指数、纹理指数以及融合植被指数和纹理指数的柑橘冠层氮素反演模型,并比较融合植被指数和纹理指数的支持向量机(SVM)、BP神经网络算法(BP)和RF的模型反演精度。结果显示:在随机森林算法中,融合植被指数和纹理指数比单独的植被指数或纹理指数更能准确预测柑橘冠层氮素含量;植被指数训练集R^(2)为0.710,测试集R^(2)为0.430;纹理指数训练集R^(2)为0.761,测试集R^(2)为0.349;融合植被指数和纹理指数训练集R^(2)为0.775,测试集R^(2)为0.533。融合植被指数和纹理指数在SVM算法训练集R^(2)为0.511,测试集R^(2)为0.371;BP神经网络训练集R^(2)为0.651,测试集R^(2)为0.204。用融合植被指数和纹理指数的RF模型对3种栽培模式的柑橘园进行氮素反演,得到宽行窄株小冠模式的柑橘冠层平均氮素含量最高,其次为宽行窄株篱壁模式,传统栽培模式最低,氮素含量均值分别为31.33、30.20和27.82 mg/g。结合无人机遥感与融合植被指数和纹理指数的随机森林算法能够有效预测柑橘冠层氮素含量,可为大尺度柑橘果园定量施肥提供参考。One hundred and twenty citrus trees under three cultivation patterns including wide row and narrow plant,wide row and narrow plant fence pattern and traditional cultivation were used to measure the content of nitrogen in the canopy and extract the texture index and vegetation index from the multispectral images data of UAV remote sensing to quickly and accurately obtain the content of nitrogen and spatial distribution characteristics of plant canopy,and to manage the large-scale orchard accurately and dynamically.The random forest(RF)algorithm was used to establish the inversion model of nitrogen in the citrus canopy based on vegetation index,texture index,and the integration of vegetation index and texture index.The inversion accuracy of support vector machine(SVM),BP neural network algorithm(BP),and RF models that integrate vegetation index and texture index was compared.The results showed that the integration of vegetation index and texture index predicted the content of nitrogen in citrus canopy more accurately than the single vegetation index or texture index in the random forest algorithm.The training sets R^(2) and the test sets R^(2) of the vegetation index,texture index,and integration of vegetation index and texture index were0.710 and 0.430,0.761 and 0.349,0.775 and 0.533,respectively.The training sets R^(2) and the test sets R^(2) of the integration of vegetation index and texture index in the SVM algorithm and BP neural network were0.511 and 0.371,0.651 and 0.204,respectively.The results of using the RF model of vegetation index and texture index to inverse the content of nitrogen in citrus orchards under three cultivation patterns showed that the average content of nitrogen in citrus canopy in wide row and narrow plant werethe highest,followed by the wide row and narrow plant fence pattern,and the traditional cultivation pattern was the lowest,with the average content of nitrogen being 31.33,30.20,and 27.82 mg/g,respectively.It is indicated that the random forest algorithm combining UAV remote sensi

关 键 词：无人机遥感 柑橘 精准施肥 氮素 多光谱 植被指数 纹理指数 

分 类 号：S127[农业科学—农业基础科学] S666