应用间隔随机蛙结合连续投影算法检测橡胶树叶片氮含量  

Application of Interval Random Frog Combined with Sucessive Projections Algorithm to Detecting Nitrogen Content in Rubber Tree Leaves

作　　者：陈伟 李创[1] 唐荣年[1] CHEN Wei;LI Chuang;TANG Rongnian(Mechanical & Electrical College,Hainan University,Haikou 570228,China)

机构地区：[1]海南大学机电工程学院

出　　处：《河南科技大学学报（自然科学版）》2019年第5期51-56,M0005,共7页Journal of Henan University of Science And Technology:Natural Science

基　　金：国家自然科学基金项目(31460318);海南省重点研究开发基金项目(ZDYF2018026);海南大学科研启动基金项目(KYQD(ZR)1872)

摘　　要：研究了基于可见/近红外光谱技术的橡胶树叶片氮含量无损检测方法。总共采集了176个橡胶树叶片样本,以350~2 500 nm的可见/近红外数据作为试验数据。首先,采用一种新颖的间隔随机蛙(iRF)算法选择交互验证均方根误差(RMSECV)最小的波长间隔组合;然后,采用连续投影算法(SPA)进一步选择特征波长;最后,将选出的特征波长作为输入变量,建立偏最小二乘法(PLS)模型。研究结果表明:通过两步策略进行波长选择,iRF算法粗选,从全光谱2 151个变量中筛选出714个,再采用SPA算法细选,从714个变量中进一步筛选出20个最优变量,降低了信息冗余,变量数减少了99.07%。建立的简化模型结果校正均方根误差(RMSEC)、校正相关系数(Rc)、预测均方根误差(RMSEP)、预测相关系数(Rp)分别为0.091 3%、0.956 5、0.123 8%和0.901 8,比PLS、间隔偏最小二乘法(iPLS)、iRF算法和SPA有更低的均方根误差和更高的相关系数。因此,iRF-SPA可以作为一种策略的波长选择方法用于检测橡胶树叶片氮含量。The non-destructive detection method of nitrogen content in rubber tree leaves based on visible/near infrared spectroscopy was studied.A total of 176 rubber tree leaf samples were collected,and the visible/near infrared data from 350~2 500 nm was used as experimental data.Firstly,a novel interval random frog(iRF)algorithm was used to select the wavelength interval combination with smallest root mean square error of crossvalidation(RMSECV).Secondly,the successive projections algorithm(SPA)was used to further select the characteristic wavelength.Thirdly,the selected characteristic wavelength was taken as the input variable to establish a partial least squares(PLS)model.The results show that the wavelength selection is carried out by two-step strategy.The 714 variables from full-spectrum 2 151 are selected by rough selection of iRF algorithm.Then the 20 optimal variables further was selected from the 714 variables refined selection by SPA algorithm,which reduces the information redundancy,and significantly reduces the number of variables of 99.07%.The root mean square error of calibration(RMSEC),correlation coefficient of correlation(Rc),root mean square error of prediction(RMSEP)and correlation coefficient of prediction(Rp)are respectively 0.091 3%,0.956 5,0.123 8%,and 0.901 8 which have lower root mean square errors and higher correlation coefficients than PLS,iPLS,iRF,and SPA.Therefore,iRF-SPA can be used as the strategy for wavelength selection to detect the nitrogen content of rubber tree leaves.

关 键 词：橡胶树叶片 氮含量 波长选择 间隔随机蛙算法 连续投影算法 

分 类 号：S794.1[农业科学—林木遗传育种]