基于机器学习与光谱技术的油菜叶片含水率估测研究

Estimation of water content in rape leaves by spectral reflectance combined with machine learning

油菜叶片的含水率是油菜的光合作用的重要因素之一,快速准确检测油菜叶片的含水率对有效诊断作物水分状况、高效利用水资源提高作物品质以及产量具有重要意义。本试验以油菜叶片为实验样品,采用烘干箱对叶片进行去水分处理,在各采集过程利用RSR-1100光谱仪获取相对应的光谱信息,同时计算出叶片的实际含水率。为了降低干扰以及消除噪声,采用标准正态变量法、SG卷积平滑滤波法、多元散射校正、均值归一化与正交信号校正等方法对光谱数据进行预处理。采用主成分分析对数据进行降维处理,最终选择多元线性回归、偏最小二乘回归、支持向量回归建立18个估测模型。结果表明,经过正交信号校正预处理后的光谱数据结合主成分分析建立的支持向量回归模型具有最优的估测效果,其决定系数Rc^(2)与Rv^(2)分别高达0.901和0.857,RMSE低至9.874%,RPD高至2.929。因此,利用光谱分析技术估算油菜的含水率是可行的,本研究可为精确灌溉和有效节水提供理论依据。

Water is one of the important components in the growth and development of rape.Rapid and accurate detection of water content in cabbage leaves is of great significance to effectively diagnose the water status of crop and improve crop quality and yield by efficient use of water resources.In this paper,rape leaves are used as experimental samples,and the leaves are eliminated moisture by oven drying method.Spectrometers RSR-1100 are used to obtain spectral information in each drying processing,and the actual moisture content of the leaves is collected.In order to reduce interference and eliminate noise,standard normal variate,Savitzky-Golay smoothing,multiplicative scatter correction,normalization and orthogonal signal correction are used to preprocess spectral data.The data is dimensionality reduction processed by principal component analysis,and finally,multiple linear regression,partial least squares regression and support vector regression are selected to establish 18 models.The results show that the support vector regression model based on the spectral data preprocessed by orthogonal signal correction and principal component analysis has the best estimation effect,with the coefficient of determination Rc^(2) and Rv^(2) respectively as high as 0.901 and 0.857,RMSE as low as 9.874%and RPD as high as 2.929.Therefore,spectral analysis technology is feasible to estimate the water content of rape,which provides a theoretical basis for precision irrigation and effective water conservation.