烤烟叶片氯密度高光谱预测模型的建立

Construction of Hyperspectral Prediction Model for Chlorine Density of Flue-cured Tobacco Leaves

连续2a设置烤烟3个品种处理和3个地点处理,提取10个植被指数[修正三角形植被指数(Modified triangular vegetation index,MTVI)、归一化植被指数1(Normalized difference vegetation index 1,NDVI1)、归一化植被指数2(Normalized difference vegetation index 2,NDVI2)、新型植被指数(New vegetation index,NVI)、比值植被指数1(Ratio vegetation index 1,RVI1)、比值植被指数2(Ratio vegetation index 2,RVI2)、比值植被指数3(Ratio vegetation index 3,RVI3)、水分指数(Water index,WI)、归一化色素叶绿素植被指数(Normalized chlorophyll pigment vegetation index,NCPI)、简单比值水分指数(Simple ratio water index,SRWI)],用一元线性回归模型、多元线性回归模型、BP神经网络模型分别对烤烟叶片氯密度进行估算,比较其对烤烟叶片氯密度的预测效果。结果表明, NDVI2 、 NVI、RVI2、RVI3、NCPI、SRWI 6个植被指数与烤烟叶片氯密度均极显著相关,相关系数均> 0.680 。一元线性回归模型、多元线性回归模型、BP神经网络模型的决定系数分别为0.617、 0.617 、0.868,其均方根误差分别为1.573、1.577、0.828。BP神经网络的预测效果比一元线性回归模型、多元线性回归模型预测效果好。

Three variety treatments and three regional treatments were set up in this experiment for two consecutive years.10 vegetation indexes [modified triangular vegetation index(MTVI),normalized difference vegetation index 1(NDVI1),normalized difference vegetation index 2(NDVI2),new vegetation index(NVI),ratio vegetation index 1(RVI1),ratio vegetation index 2(RVI2),ratio vegetation index 3(RVI3),water index(WI),normalized chlorophyll pigment vegetation index(NCPI),simple ratio water index(SRWI)] were extracted and analyzed.The chlorine density of flue-cured tobacco leaves was estimated by simple linear regression model,multiple linear regression model and BP neural network model to compare the prediction effect of chlorine density of flue-cured tobacco leaves.The result showed that the 6 vegetation indexes of NDVI2,NVI,RVI2,RVI3,NCPI,SRWI were significantly correlated with the chlorine density of flue-cured tobacco leaves,and the correlation coefficients were all greater than 0.680.The determining coefficients of simple linear regression model,multiple linear regression model and BP neural network model were 0.617, 0.617 and 0.868,respectively,and the root mean square errors were 1.573,1.577 and 0.828,respectively.The prediction effect of BP neural network is better than that of simple linear regression model and multiple linear regression model.