基于高光谱成像技术的油菜叶片叶绿素含量预测

Prediction of Chlorophyll Content of Rape Leaves with Hyperspectral Imaging Technology

叶绿素是作物生长中的重要因素,可用于实时监测作物的生长状况。以常规高油酸油菜品种为材料,采用大田试验研究油菜叶片在不同栽培措施下幼苗期、蕾薹期叶片的光谱响应,通过计算反射光谱及其反射光谱的一阶导数与SPAD值的相关性,结合逐步回归挑选出油菜叶片敏感波段,并计算光谱指数。采用一元线性回归和神经网络建立叶绿素含量估算模型。结果表明,由光谱指数所构建的神经网络叶绿素估算模型,精度评价结果均显示比较高的水平,幼苗期反射率光谱指数构建的模型精度最高,决定系数R^2为0.807 0,均方根误差(RMSE)为1.131 5,蕾薹期一阶导数光谱指数构建的模型精度最高,决定系数R^2为0.873 2,均方根误差(RMSE)为1.322 3,在蕾薹期和幼苗期通过构建BP神经网络模型能够较好的对油菜叶绿素进行反演。为利用高光谱技术大范围监测油菜叶绿素含量提供了一定的理论依据。

Chlorophyll is an important factor in crop growth, which can be used to real time monitor the crop growth. Taking conventional high-oleic acid rape variety as material, this paper conducted field experiment to study the spectral response of rape leaves at seedling stage and bolting stage under different cultivation measures. And then, it calculated the correlation between the first derivative of reflectance spectrum and its reflection spectrum with SPAD value, selected sensitive bands of rape leaves combining with the stepwise regression and computed the spectral index. The estimation model of chlorophyll content was established by linear regression and neural network. The results showed that the accuracy of the neural network chlorophyll estimation model constructed by spectral index showed relatively high level, and the model constructed by the spectral indices of seedling stage had the highest precision. The coefficient of determination R^2 was 0.807 0, the root mean squared error(RMSE) was 1.131 5. The model accuracy constructed by the first order derivative spectral indices of the bolting stage was the highest. The coefficient of determination R^2 was 0.873 2, the root mean squared error(RMSE) was 1.322 3. Inversion of rape chlorophyll by constructing BP neural network model was better during the seedling and bolting stage. Above results provided a theoretical reference for monitoring the chlorophyll content of rapeseed in a wide range by hyperspectral technology.