基于高光谱和连续投影算法的猕猴桃叶片氮平衡指数的估测  

Estimation of kiwifruit leaf nitrogen balance index based on hyperspectral and successive projections algorithm

作　　者：宋子怡 常庆瑞[1] 郑智康 唐国强 孟怡凡 SONG Ziyi;CHANG Qingrui;ZHENG Zhikang;TANG Guoqiang;MENG Yifan(College of Natural Resources and Environment,Northwest A&F University,Yangling 712100,China)

机构地区：[1]西北农林科技大学资源环境学院,陕西杨凌712100

出　　处：《江苏农业学报》2024年第7期1260-1267,共8页Jiangsu Journal of Agricultural Sciences

基　　金：国家高技术研究发展计划资助项目(2013AA102401)。

摘　　要：通过研究猕猴桃叶片氮平衡指数(Nitrogen balance index,NBI)与高光谱反射率之间的关系,建立合适的遥感估算模型,以期为指导陕西咸阳地区猕猴桃生长监测及田间精准施氮奠定理论基础。以陕西省咸阳市杨凌区的徐香猕猴桃为主要研究对象,测定其高光谱反射率、叶片氮平衡指数,通过一阶导数、二阶导数、连续统去除和标准正态分布光谱变换,分析包含原始光谱在内的5种不同光谱与叶片氮平衡指数之间的关系;进一步通过连续投影算法,剔除冗余信息,筛选出特征波长,并基于不同光谱的特征波长,使用单因素回归模型、随机森林回归(Random forest regression,RF)模型、支持向量机回归(Support vector regression,SVR)模型和偏最小二乘回归(Partial least square regression,PLSR)模型进行建模,比较模型精度。结果表明,当NBI值不同时,猕猴桃叶片相关指标的变化趋势类似:可见光波段的反射率随NBI值的增加呈现下降的趋势,而近红外波段反射率的变化趋势则与之相反,表现出随NBI值的增加而上升的趋势;部分光谱变换可以增加通过0.01水平显著性检验的波段数,提升与NBI值的相关性,其中连续统去除光谱的敏感波段数增加得最多,增加了190个,一阶导数光谱相关系数的绝对值最大值为0.77;连续投影算法可最大限度地减少数据的冗余,最高降维比达99%,在大幅提高计算效率的同时提高了模型的精度;与单因素回归模型相比,多因素机器学习模型对猕猴桃氮平衡指数的估算能力较高,其中SNV-SVR的表现最好,决定系数(R 2)为0.82,相对百分比差异(RPD)为2.34。在今后对猕猴桃氮平衡指数的估测中,可优先考虑本研究模型。By investigating the relationship between kiwifruit leaf nitrogen balance index(NBI)and hyperspectral reflectance,this study established a suitable remote sensing estimation model to provide a theoretical foundation for guiding precise nitrogen management and growth monitoring of kiwifruit in the Xianyang region.Taking Xuxiang kiwifruit in Yangling of Xianyang City,Shaanxi province as the main research object,the hyperspectral reflectance and leaf nitrogen balance index were measured.Through the first derivative,second derivative,continuum removal and standard normal distribution spectral transformation,the relationship between five different spectra including the original spectrum and leaf nitrogen balance index was analyzed.Furthermore,through the successive projections algorithm,the redundant information was eliminated and the characteristic wavelengths were screened out.Based on the characteristic wavelengths of different spectra,single factor regression model,random forest regression(RF)model,support vector regression(SVR)model and partial least square regression(PLSR)model were used for modeling,and the model accuracy was compared.The results showed that when the NBI value was different,the change trend of the related indices of kiwifruit leaves was similar.The reflectivity of the visible light band showed a downward trend with the increase of the NBI value,while the change trend of the reflectivity of the near-infrared band was opposite,showing an upward trend with the increase of the NBI value.Partial spectral transformation increased the number of bands passing the 0.01 level significance test and improved the correlation with NBI values.The number of sensitive bands of continuum removal spectra increased by 190,and the maximum absolute value of the first derivative spectral correlation coefficient was 0.77.The successive projections algorithm could minimize the redundancy of data,and the highest dimensionality reduction ratio was as high as 99%.It greatly improved the computational efficiency and the ac

关 键 词：猕猴桃 叶片氮平衡指数 高光谱 光谱变换 连续投影算法 支持向量机回归 

分 类 号：S127[农业科学—农业基础科学]