高光谱影像的苹果花叶病叶片花青素定量反演

花叶病是苹果叶片常见的病毒性病害，患病叶片的花青素含量出现异常。以叶片花青素含量作为病害严重程度的定量化指标，使用高光谱成像技术获取感染花叶病的苹果叶片的高光谱图像，分析叶片的光谱特征，通过任意两个波段的反射率的不同数学组合，构建并筛选对染病叶片花青素含量高度敏感的最优光谱指数，进而建立苹果叶片花青素含量的高光谱估算模型，最终实现苹果叶片花青素含量分布状况的可视化表达。结果表明，随着病害严重程度的增大，苹果叶片的花青素含量升高；叶片染病区域的光谱反射率在整个可见光区域明显增加，而且出现了红边蓝移现象。通过两两波段组合构建的三种光谱指数(NDSI_(770，722)，RSI_(717，770)，DSI_(581，520))与苹果叶片花青素含量的相关系数绝对值均达到0.8以上。在构建的四种苹果叶片花青素含量估算模型中，选用三个光谱指数为参数、并使用偏最小二乘回归方法建立的Anth-PLSR模型精度最高(R2=0.823, RMSE=0.056)。采用Anth-PLSR模型对患病叶片的高光谱图像进行逐像元解算，得到苹果花青素含量分布图。进一步通过叶片花青素含量分布图计算苹果叶片整叶的花青素含量平均值，作为苹果叶片健康程度的定量化指标。此外，通过提取整叶光谱均值、使用同样模型可简洁有效地估算苹果整叶花青素含量平均值。为苹果叶片花叶病病害监测提供了一种直观、快速的技术手段。

Quantified Estimation of Anthocyanin Content in Mosaic Virus Infected Apple Leaves Based on Hyperspectral Imaging

Anthocyanin has the antioxidant effect,which is helpful to the recover of leaf injury.The dynamic change of anthocy-anin concentration can be considered as a sensitive indicator to reflect plant physiological conditions affected by external environ-mental stresses,consequently the anthocyanin content of mosaic virus infecting apple leaves can be used as an important criterion for assessing the degree of disease.In this research,hyperspectral images of apple leaves with mosaic disease were acquired by imaging spectrometer.By the combination of each two bands,spectral reflectance was used to establish the optimal spectral inde-xes which were highly correlated to the anthocyanin content in infected leaves.Further more,a accurate anthocyanin concentra-tion estimation model was established taking these spectral indexes as parameters.The results were as follows:(1)The damage of mesophyll cells in mosaic virus infecting apple leaves would cause the increase of anthocyanin content.As a result,the spec-tral reflectance of infected area increased significantly in visible region,while in near infrared region the reflectance was lower than the normal.In addition,the red-edge position shifted to the shorter wavelength and both the red-edge area and the first de-rivative spectral reflectance at red-edge position decreased.(2)The correlation between leaf anthocyanin content and spectral re-flectance was extremely significant at most of the wavebands and reached the peak at 581 nm.The normalised deviation spectral index combined with spectral reflectance at 770 and 722 nm,the simple ratio spectral index combined with spectral reflectance at 717 and 770 nm and the deviation spectral index correlated with spectral reflectance at 581 and 520 nm were all significantly relat-ed to leaf anthocyanin content,with the correlation coefficent of 0.838,0.865 and 0.875,respectively.(3)Anth-PLSR was the optimal model to estimate apple leaf anthocyanin content,of which the determination coefficient was 0.823 and RMSE was 0.056.The anthocyanin content distribution diagram of leaves were made by solving the hyperspectral images pixel using Anth-PLSR model,thus the anthocyanin content of an integral leaf was calculated.On the other hand,by extracting the average spec-tral reflectance from the the hyperspectral image of a whole leaf,the anthocyanin content of the integral leaf can be obtained using Anth-PLSR model.The results of these two different methods showed a high consistence by fitting analysis,which dem-onstrated that the latter method could be used to rapidly detect the anthocyanin content of apple leaves.