摘要
为实现快速无损检测竹叶片氮含量,采用波长范围为350~2 500 nm的地物光谱仪获取竹叶片光谱数据,以金镶玉竹叶片为样本,对其进行高光谱分析。将高光谱原始反射率及其一阶微分、对数一阶微分和二阶微分值,与化学法测量的竹叶片氮含量值进行了相关性分析,分别获得了不同微分变化下的特征波段;基于微分变换后的高光谱反射率数据,分别采用二元线性回归、多元逐步回归、偏最小二乘回归和基于主成分分析的BP神经网络方法,建立了4种金镶玉竹叶片的氮含量高光谱估测模型。对比4种估测模型的校验结果表明,在光谱反射率的对数一阶微分变换下,采用拓扑结构为6-10-1的基于主成分分析的BP神经网络估测模型,校验环节决定系数为0. 838,均方根误差RMSE为0. 045 2,具备较好的竹叶片氮含量估测效果。
In order to achieve rapid and non-destructive detection of nitrogen content in bamboo leaves,Phyllostachys aureosulcata leaves were used as samples for the hyperspectral analysis. To a certain extent,the nitrogen content in plant leaves can reflect the nitrogen condition inside the plant,which has a good prediction effect on plant growth. The spectral data of bamboo leaf was obtained by using the field portable terrain spectrometer with spectral range from 350 nm to 2 500 nm. Correlational analysis was conducted between the nitrogen content measured by the chemical method and the hyperspectral reflectance,the first order differential reflectance,the logarithmic first order differential reflectance and the second order differential reflectance of bamboo leaves,respectively,and the characteristic bands were obtained. Four estimation models of nitrogen content of bamboo leaf were established by the binary linear regression,multivariate stepwise regression,partial least squares regression(PLRS) and principal component analysis-BP neural network regression(PCA-BP). respectively. The experimental results of four estimation models showed that by using the logarithmic first order difference of the hyperspectral reflectance,PCA-BP estimation model with 6-10-1 topology based on principal component analysis had better estimation result. The determination coefficient(R2) and root mean square error(RMSE) were0. 838 and 0. 045 2,respectively.
出处
《农业机械学报》
EI
CAS
CSCD
北大核心
2018年第S1期393-400,共8页
Transactions of the Chinese Society for Agricultural Machinery
基金
中央高校基本科研业务费专项资金项目(2015ZCQ-GX-04)
北京市科技计划项目(Z161100000916012)
北京市共建项目
关键词
竹叶片
氮含量
高光谱
二元线性回归
多元逐步回归
偏最小二乘回归
bamboo leaf
nitrogen content
hyperspectra
binary linear regression
multiple stepwise regression
partial least squares regression
