结合冠层光谱和叶片生理观测的小麦条锈病监测模型研究

通过开展小麦条锈病接种试验,在多个关键生育期获取被动式的冠层光谱和主动式的叶片生理观测并开展病情调查。在此基础上,结合优选的光谱特征和生理特征采用偏最小二乘回归方法(PLSR)构建病情严重度反演模型,得到不同生育期精度表现最优的特征组合。结果显示,基于光谱观测的优选光谱特征和基于叶片生理观测的Flav(类黄酮相对含量)、Chl(叶绿素含量)的不同组合在小麦挑旗期、灌浆早期和灌浆期分别具有较佳表现,模型精度达到r~2=0.90,RMSE=0.026。相比单纯采用光谱特征,综合冠层光谱和叶片生理观测能够使模型精度提高21%,表明两种数据的结合有利于提高病情严重度估测精度。上述研究可为小麦病害监测仪器的开发提供新的模式和思路。

Combined canopy spectral and leaf physiological observations in model development for wheat stripe rust detection

This study attempted to combine measurements from both passive and active sensors to form a retrieving model of wheat stripe rust severity. In a disease inoculation experiment, besides the survey of disease severity, measurements of both the passive canopy spectra and active foliar fluorescence were carried out at two key growing stages. Prior to model development, a feature selection protocol is implemented to identify optimal features serving as model input variables. Based on different combinations of the selected features, the retrieving models of disease severity were developed and compared using the partial least squares regression (PLSR) method, to determine the best feature combinations at different growing stages. The results based on the optimal spectral features and leaf physiological observations on Flav (flavonoids), Chl (chlorophyll) of different combinations at wheat flag, early filling and grain filling stages had a better performance, with a precision of r2=0.90, and RMSE=0.026. Compared to spectral characteristics alone, comprehensive canopy spectra and leaf physiological observations improved model accuracy by 21%, showing that the combination of the two kinds of data could improve the disease severity estimation precision. The study can provide a new pattern and idea for the development of wheat disease monitoring instrument.