基于分数阶微分光谱指数的小麦条锈病遥感监测模型构建

为提高小麦条锈病的遥感监测精度，该研究利用分数阶微分能够突出光谱的细微信息以及描述光谱数据间微小差异的优势，在对条锈病胁迫下小麦冠层光谱数据进行分数阶微分处理的基础上，构建了两波段和三波段分数阶微分光谱指数，并将其应用于小麦条锈病的遥感探测。研究结果表明，1.2阶次微分光谱与小麦条锈病冠层病情严重度的相关性最高，较原始反射率光谱、一阶微分光谱和二阶微分光谱分别提高了20.9%、3.9%和20.5%；基于分数阶微分光谱指数的最优分数阶次及其对应波长构建的三波段分数阶微分光谱指数对小麦条锈病的探测能力优于两波段分数阶微分光谱指数，其中分数阶微分光化学指数与冠层病情严重度的相关系数达到0.875；以分数阶微分光谱指数为自变量构建的高斯过程回归（Gaussian Process Regression，GPR）模型对小麦条锈病冠层病情严重度的预测精度优于反射率光谱指数，其训练数据集及验证数据集病情指数（Disease Index，DI）预测值和实测值间的决定系数较反射率光谱指数分别提高了3.8%和19.1%，该研究结果对进一步实现作物健康状况大面积高精度遥感监测具有重要意义。

Construction of remote sensing monitoring model of wheat stripe rust based on fractional-order differential spectral index

Hyper spectral data is the most vulnerable to environmental noise (such as soil background) when monitoring wheat stripe rust. The first- and second-order differential processing of spectral data can be used to eliminate part of the noise, but it is easy to ignore the detailed information of stripe rust. In this study, a fractional-order differential spectral index was proposed to process the hyperspectral data of wheat canopy under the stress of stripe rust. Three two-band and three three-band fractional-order spectral differential indices were constructed after the band combination optimization, according to the current six types of spectral index. Gaussian regression was also applied to estimate the severity of stripe rust disease, compared with the commonly-used reflectivity spectral index. The results showed that the correlation between the fractional-order differential spectrum and the disease index of stripe rust was more significant than that of the original spectrum, where the most obvious significance was found in the range of 0.3-1.3 order differential spectrum. The correlation coefficient was the largest for the 481 nm band of 1.2 order differential spectrum with the severity of wheat stripe rust, 20.9%, 3.9%, and 20.5% higher than that of the original reflectance spectrum, the first-, and the second-order differential spectrum, respectively. Two-band fractional-order differential spectral indices were determined by the maximum correlation coefficient. Specifically, the values of the best order for the fractional-order differential-difference index, ratio index, and normalized difference index were 0.4, 1.3 and 1.2, respectively, where the band combination was 481 nm and 475 nm, 478 nm and 622 nm, as well as 481 nm and 673 nm, respectively. In the three-band fractional-order differential-difference index, the best order of fractional-order differential improved difference index was 1.1, and the band combination was 481, 442, and 454 nm. The best order of fractional-order differential improved ratio index was 1.2, and the band combination was 880, 670, and 481 nm. The best order of fractional-order differential photochemical reflectance index was 0.5, and the band combination is 646, 400, and 955 nm. The correlation between the three-band fractional-order differential spectral index and the severity of wheat stripe rust was better than that of the two-band fractional-order differential spectral index, where the fractional-order differential photochemical reflectance index presented the highest correlation with the severity of wheat stripe rust. Furthermore, the Gaussian regression model using the fractional-order differential spectral index indicated a better prediction accuracy for the stripe rust disease index than that for the reflectance spectral index. The determination coefficient between the predicted and measured values of Disease Index (DI) in the training and validation data set increased by 3.8% and 19.1%, respectively, where the Root Mean Square Error (RMSE) decreased by 13.0% and 33.5%, respectively, compared with the reflectance spectral index. Consequently, the fractional-order differential spectral index can be expected to improve the remote sensing detection accuracy of wheat stripe rust. This finding can provide a promising feasible way for the hyper spectral remote sensing to monitor the wheat stripe rust, thereby to realize the large-scale high-precision remote sensing monitoring of crop health.