基于支持向量机的棉花冠层叶片叶绿素含量高光谱遥感估算  

Hyperspectral estimation of chlorophyll content of cotton canopy leaves based on support vector machine

作　　者：张卓然 常庆瑞[1] 张廷龙[1] 班松涛[1] 由明明 ZHANG Zhuoran;CHANG Qingrui;ZHANG Tinglong;BAN Songtao;YOU Mingming(College of Natural Resources and Environment,Northwest A&F University,Yangling,Shaanxi 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：《西北农林科技大学学报（自然科学版）》2018年第11期39-45,共7页Journal of Northwest A&F University(Natural Science Edition)

基　　金：国家高技术研究发展计划(863计划)项目(2013AA102401-2)

摘　　要：【目的】建立并研究棉花冠层叶片叶绿素含量的高光谱估算模型,探讨合适的建模方法,以提高棉花叶绿素含量的高光谱遥感估算精度。【方法】以2016年种植的渭北旱塬区棉花鲁棉研28号为试验对象,用SPAD-502型手持式叶绿素仪和HR-1024i便携式地物光谱仪,分别测定棉花不同生育期冠层叶片SPAD值和对应的光谱反射率,分析SPAD值与光谱反射率的相关性。选取8个光谱参数,分析SPAD值与这8个光谱参数的相关性,并采用单因素回归、多元逐步回归和支持向量机(SVM)回归方法,构建棉花冠层叶片叶绿素含量的高光谱估算模型,比较各模型的决定系数(R2)、均方根误差(RMSE)以及相对误差(RE),评价模型的精度。【结果】(1)棉花冠层叶片光谱反射率在400～700nm波段随叶片SPAD值升高而降低,在700～1 000nm波段表现为SPAD值越高,叶片光谱反射率越高;(2)在530～570nm和680～730nm处叶绿素含量与光谱反射率呈极显著负相关(99.99%置信区间,n=144);(3)所选用的8个光谱参数与叶绿素含量均达到极显著相关,相关系数最高为0.686;(4)SVM回归模型验证R2达到了0.884,RMSE和RE最低,分别为2.186和3.419,比单因素回归模型中预测精度最高的SPAD-RVI1的RMSE和RE分别降低46.4%和46.3%,较多元逐步回归模型SPAD-MSR的RMSE和RE分别降低33.4%和32.1%,明显提高了棉花叶绿素含量的估算效果。采用8个光谱参数构建的SPAD-SVM8模型RMSE和RE比采用4个光谱参数构建的SPAD-SVM8模型分别降低了19.2%和23.5%。【结论】支持向量机(SVM)回归方法可以作为棉花冠层叶片叶绿素含量高光谱遥感估算的优选方法,且采用较多光谱参数构建的SVM模型估算精度更高。【Objective】This study established and investigated hyperspectral estimation models for chlorophyll content of cotton canopy leaves and explored appropriate modeling method to improve hyperspectral estimation of cotton chlorophyll content.【Method】Lumianyan 28 planted at Weibei dry highland in 2016 was selected and the chlorophyll content of cotton leaves was estimated with hyperspectral technology.The correlation between SPAD values obtained by SPAD-502 portable chlorophyll analyzer and spectral reflectivity obtained by SVC HR-1024i full band spectrometer was analyzed.The relationship of SPAD values and eight selected hyperspectral indices based on the sensitive wave bands were also analyzed and the cotton chlorophyll content estimation models were established using single factor polynomial linear regression method,multiple stepwise regression(MSR)method and support vector machine regression(SVM)method.Coefficient of determination(R 2),root mean square root(RMSE)and relative error(RE)were used to evaluate the models.【Result】①The spectral reflectance of cotton canopy leaves decreased with the increase of SPAD in the 400-700 nm band,while increased with the increase of SPAD in the 700-1 000 nm band.②The chlorophyll contents of cotton leaves had significantly negative correlation with spectral reflectivity at the green band(530-570 nm)and red band(680-730 nm).③The eight hyperspectral indices were significantly correlated with chlorophyll content with the maximum correlation coefficient of 0.686.④The R 2,RMSE and RE of eight hyperspectral indices based on SVM regression model were 0.884,2.186 and 3.419.The RMSE and RE of SPAD-SVM 8 were 46.4%and 46.3%lower than the single factor polynomial linear regression method and 33.4%and 32.1%lower than the multiple stepwise regression(MSR)method.The RMSE and RE of SPAD-SVM 8 model compare with SPAD-SVM 4 model reduce 19.2%and 23.5%.【Conclusion】Compared with other regression methods,the SVM method was the best algorithm for estimating chlorophyll content

关 键 词：棉花 叶绿素含量 SVM 光谱参数 高光谱遥感 估算模型 

分 类 号：S562[农业科学—作物学]