基于光谱分形维数的水稻白叶枯病害监测指数研究

针对缺乏有效监测水稻叶片感染白叶枯病害光谱指数的问题，以分蘖期的水稻叶片为研究对象，采集了接种白叶枯病菌的水稻叶片和对照处理的水稻叶片各200片，利用高光谱成像装置获取373~1033nm波段的水稻叶片光谱数据，选取450~900nm波段的水稻叶片高光谱数据作为样本。从每个样本中选取一个感兴趣区域（Region of interest, ROI）并计算平均光谱，经过Savtzky-Golay平滑处理得到平均光谱曲线；为了定量描述水稻叶片是否感染病害，提出将光谱分形维数(Fractal dimension, FD)作为定量描述水稻白叶枯病害的监测光谱指数，实现对白叶枯病害的监测。通过分析光谱指数(Spectral index, SI)和FD，建立SI和FD之间的多元线性关系，同时比较了FD与其他常用监测指数对白叶枯病害监测的有效性。结果表明：水稻白叶枯病害在绿峰（510~560nm）和红谷（650~690nm）波谱内的响应较为敏感；针对健康和感病叶片，FD与SI之间存在较好的多元线性关系，说明FD与光谱曲线有较好的对应关系，可以作为定量描述叶片健康状况的光谱指数；与常用监测指数相比，本文病害监测指数与水稻染病具有更高的相关性，其相关系数达到了0.9840，指数分布稳定性更高。本研究结果说明基于光谱反射曲线的圆规分形维数对判断水稻叶片是否感染白叶枯病害是可行的，为水稻白叶枯病害的监测提供了一种新方法。

Monitoring Index of Rice Bacterial Blight Based on Hyperspectral Fractal Dimension

With the rapid development of rice phenotype research, rice disease research has also made significant progress as an essential part of rice phenotype research. Bacterial blight disease is one of the three major diseases of rice. Still, there is a lack of an effective spectral index for monitoring whether rice leaves are infected with bacterial blight. Taking rice leaves at the tillering stage as the research object, totally 200 pieces of rice leaves inoculated with Xanthomonas oryzae, and control group were collected respectively. A hyperspectral imaging device was used to obtain the spectral data of rice leaves in the band of 373~1033nm, and eventually, the band of 450~900nm was selected. The hyperspectral data of rice leaves in the wave band was used as a sample. The region of interest (ROI) was selected from each sample and the average spectrum was calculated. After applying Savtzky-Golay smoothing, the average spectrum curve was obtained. In order to quantitatively describe whether the rice leaves were infected or not, the spectral fractal dimension (FD) as a monitoring spectral index for quantitatively describing rice bacterial leaf blight disease was used. By analyzing the spectral index (SI) and FD, the multivariate linear relationship between SI and FD was established, and the effectiveness of FD and other commonly used monitoring indexes for bacterial blight monitoring were compared. The results showed that the response of rice bacterial leaf blight in the green peak (510~560nm) and red valley (650~690nm) spectrum was more sensitive;for healthy and susceptible leaves, there was a good relationship between FD and SI. The multivariate linear relationship of FD indicated that FD had a good corresponding relationship with the spectral curve, which can be used as a spectral index to quantitatively describe the health of leaves;compared with the commonly used monitoring index, the proposed disease monitoring index had a high correlation with whether rice was infected or not. The correlation coefficient reached 0.9840, and the distribution was more stable. The results indicated that the fractal dimension based on the spectral reflectance curve was feasible for judging whether rice leaves were infected with bacterial blight and provided a method for early monitoring of rice bacterial blight.