Wheat powdery mildew (WPM) mainly damages leaves, which leads to yield reduction. Although spectroscopy has been considered as one of the effective methods to detect crop diseases, the severity estimation of WPM is challenged by complex changes of wheat canopy reflection, wavelength redundancy in hyperspectral data et al., so that to reduce the accuracy of disease index (DI) estimation. Aiming to estimate DI of WPM, an optimization method combined of band intervals and sensitive wavelengths screening was proposed by in-situ reflected spectral responses. Field experiments were conducted in 2022 and 2023. With regards to data processing, after preprocessing by Savitzky-Golay (SG) and multiplicative scatter correction (MSC) methods, the backward interval partial least squares (Bipls) method was used to featuring band interval. Then, the characteristic wavelengths in the selected interval were optimized by the variable important in projection (VIP), maximal information coefficient (MIC) and variable importance measure based on random forest (RFVIM) method, separately. Finally, partial least squares regression (PLSR) was used to establish the DI estimating model of WPM, and the best model was selected by comparing the performance among the models. The results indicate that the Bipls-VIP-PLSR model has the best performance (2022: = 0.78, RMSEV = 10.94%; 2023: = 0.75, RMSEV = 11.73%), and it was used to estimate the DI changes in the field. The distribution map of DI provided a potential for WPM monitoring and plant protection decision-making.

中文翻译：

---

基于原位高光谱响应和特征波长优化的小麦冠层白粉病估算


小麦白粉病(WPM)主要危害叶片，导致减产。虽然光谱学被认为是检测农作物病害的有效方法之一，但WPM的严重程度估计受到小麦冠层反射的复杂变化、高光谱数据中波长冗余等的挑战，从而降低了病害指数的准确性。 DI）估计。针对WPM DI的估计，提出了一种结合波段间隔和敏感波长筛选的原位反射光谱响应的优化方法。 2022年和2023年进行了现场实验。数据处理方面，经过Savitzky-Golay（SG）和乘性散射校正（MSC）方法预处理后，采用后向间隔偏最小二乘（Bipls）方法来表征波段间隔。然后，分别通过投影重要变量（VIP）、最大信息系数（MIC）和基于随机森林的变量重要性测度（RFVIM）方法对选定区间内的特征波长进行优化。最后，利用偏最小二乘回归（PLSR）建立WPM的DI估计模型，并通过比较模型之间的性能来选择最佳模型。结果表明，Bipls-VIP-PLSR模型具有最佳性能（2022年：= 0.78，RMSEV = 10.94％；2023年：= 0.75，RMSEV = 11.73％），用于估计现场DI变化。 DI的分布图为WPM监测和植保决策提供了可能。