基于高光谱特征波长的冬小麦水分含量估测模型

Study on Water Content Estimation Model of Winter Wheat Based on Hyperspectral Characteristic Wavelength

【目的】以高光谱技术为核心,结合理化数据,建立快速、无损的冬小麦冠层水分含量估算模型,为利用高光谱技术进行小麦水分含量的无损检测提供参考。【方法】测定两种冬小麦的叶片、植株含水量,采集其光谱数据作SG平滑、一阶导数和二阶导数处理,分析其相关关系,构建冬小麦叶片和植株含水量的多种估算模型,进行精度评价。【结果】不同光谱数据处理中一阶导数变换能够显著增加与小麦含水量的相关性,叶片含水量在456 nm波长处达到了最大负相关,相关系数为0.87,植株含水量在457 nm波长处达到了最大负相关,相关系数为0.8909;偏最小二乘回归构建的水分含量估测模型拟合精度优于线性和多元回归模型,线性模型采用R 650、SG 1944、R′456、R″681构建的模型估测叶片含水量较好,估测植株含水量R 664、SG 663、R′457、R″681精度较高;多元线性回归和偏最小二乘回归都是采用一阶导数变换构建的模型拟合精度最高,叶片和植株水分含量估测模型的外部检验R 2分别达到0.8032、0.8670、0.8540、0.8856。【结论】小麦原始光谱一阶导数变换后能够显著提高与水分含量的相关性,利用PLSR方法构建的小麦水分含量估测模型拟合精度最高。

【Objective】To establish a rapid and non-destructive canopy moisture content estimation model of winter wheat based on hyperspectral technology and physicochemical data in the hope of providing a theoretical basis for the rapid monitoring of winter wheat.【Method】The water content of leaves and plants of two kinds of winter wheat samples were measured,the spectral data were collected for SG smoothing,first-order derivative and second-order derivative processing,and the correlation between the three was analyzed.And finally,multiple estimation models of the water content of leaves and plants of winter wheat were constructed and the accuracy was evaluated.【Results】The first-order derivative transformation in different spectral data processing could significantly increase the correlation with the water content of wheat.The water content of leaves reached the maximum negative correlation at the wavelength of 456 nm,and the correlation coefficient was 0.87.The water content of plants reached the maximum negative correlation at the wavelength of 457 nm,and the correlation coefficient was 0.890,9.The fitting accuracy of the water content estimation model constructed by partial least square regression was better than that of the linear and multiple linear regression models.The linear model using the model constructed by R 650,SG 1944,R′456 and R″681 was better to estimate the water content of leaves,and the accuracy to estimated plant water content of R 664,SG 663,R′457 and R″681 was higher.Multiple linear regressions and partial least squares regression were both models constructed by using the first derivative transformation,and both of them had the highest fitting accuracy.The external test R 2 of the leaf and plant water content estimation models reached 0.803,2,0.867,0,0.854,0 and 0.885,6,respectively.【Conclusion】The first-order derivative transformation of the original spectrum of wheat can significantly improve the correlation with the moisture content.The estimation model of wheat water content constructed by PLSR method has the highest fitting accuracy,which provides a reference for the non-destructive detection of wheat moisture content by using hyperspectral technology.