连续投影算法在小麦高光谱定量分析中的应用.pdfVIP

电子技术研发 Electronics R D 连续投影算法在小麦高光谱定量分析中的应用 王 劼 秦琳琳 吴 刚 （中国科学技术大学自动化系） 摘 要：为了消除光谱变量间的共线性影响，减少建模变量以提高校正速度，文章引进一种新的波长选择方法： 采用连续投影算法(Successive Projections Algorithm，SPA)，并将其集成偏最小二乘(Partial Least Squares，PLS)技术构 成SPA-PLS方法，用于冬小麦高光谱波长优化选择及其与叶绿素含量的定量分析。对119个小麦叶片样本的实验结果表 明，在经SPA-PLS算法后，优选了全波段的7个特征波长（占全谱的1.173％）来建立预测模型，交互验证均方根偏差 (RMSECV) 由未经过波长选择时的1.4117降至0.7893，模型相关系数r为0.6737，建模均方根误差RMSEC为0.7508 mg/g ，预测均方根误差RMSEP为0.9701mg/g 。 关键词：高光谱；波长选择；连续投影算法；PLS ；定量分析 Application of Successive Projections Algorithm in Quantitative Analysis of Winter Wheat Hyper-spectral Variables Wang Jie Qin Linlin Wu Gang （Department of Automation, University of Science Technology of China ） Abstract: In order to minimize the co-linearity effects of variables in the calibration data set, reduce the modeling variables to alleviate the computation workload, successive projections algorithm (SPA) is applied to extract feature wavelengths to optimize hyper-spectral calibration model in this study. Applying the proposed method to 119 winter wheat samples, 7 variables (account for 1.17% of all variables) are selected by SPA to build PLS model in calibration set, the root mean square error of cross validation(RMSECV) is reduced from 1.4117 to 0.7893. The experimental results show that the calibration correlation coefficient (r) is 0.6737, and RMSEC (the value of the root-mean-square error of calibration) is 0.7508 mg/g, RMSEP (the value of the root-mean-square error of prediction) is 0.9701mg/g. Key words: hyper-spectral; wavelength selection; successive projections algorithm; PLS; quantitative analysis 0 引言 最终选择出那些包含最小冗余信息量的变量，解决共线性