基于近距离成像高光谱的水稻叶瘟病早期探测机理与方法研究

Rice blast is the most devastating disease with frequent occurrences in all major rice production regions of China. The early detection of this disease with hyperspectral remote sensing is crucial for managing prevention practices and reducing yield loss. Previous studies focused on the spectral discrimination of disease levels after the emergence of visible symptoms, but did not pay much attention to the dynamics of spectral responses to the disease at the early stage. The mechanism and methods are scarce on the early detection of rice blast with imaging spectroscopy. To address these issues, we propose this project based on a greenhouse experiment with artificial inoculation of the fungal pathogen and plan to carry out systematic studies in five aspects with intensive analyses of hyperspectral imagery of rice leaves. The research contents include the leaf-level spectral responses to the disease and physiological foundation, the determination of disease sensitive spectral features from the central pixels of disease spots, the spatial and temporal variation in the spectral properties of visible and pre-visual spots, the influence of disease stress on the retrieval of biochemical parameters and the within-leaf radiative transfer process, and the discrimination of healthy and diseased leaves based on within-leaf anomaly ratios of biochemical and spectral parameters. The outcome of this project will be to establish a suite of theory and methodology for the early detection of rice leaf blast with close-range imaging spectroscopy data, and to build robust models for discriminating healthy and diseased leaves based on the fusion of within-leaf spectral and spatial information. By taking advantage of those leaf hyperspectral images, we expect to reveal the mechanism of the spectral responses to rice leaf blast at leaf and pixel scales and to make significant contributions to the theory and methods on the early detection of rice diseases. This project will provide important support to the dedicated effort on developing a robust and operational system for the remote and timely detection of rice diseases in the future.

稻瘟病是全国各水稻产区均有发生的最具毁灭性的病害，利用高光谱遥感技术实现病害的早期探测，对于适时防控和减少损失具有重大意义。当前研究对明显症状出现后病情等级的光谱识别关注较多，而对病害早期阶段的光谱响应特性研究不够深入，针对早期病害的成像高光谱探测机理和方法还非常缺乏。项目拟在温室开展叶瘟病菌人工接种试验，以叶片高光谱影像分析为核心，围绕叶瘟病胁迫下叶片水平光谱响应规律及其生理基础、基于中心像素的叶瘟病敏感光谱特征、叶内可见与潜在病斑光谱时空变化规律、生化参数反演及叶片辐射传输的响应特性、基于叶内生化与光谱异常率的感病叶片识别模型等五个方面开展研究。拟建立一套基于近距离成像高光谱平台的叶瘟病早期探测理论与技术，构建融合叶内光谱与空间信息的感病叶片识别模型。项目拟阐明水稻叶片对叶瘟病的光谱响应机理，在病害早期识别理论与方法上取得突破，为构建稳定高效的水稻病害监测系统提供重要技术支撑。

稻瘟病是威胁全球水稻生产的最具破坏性的真菌病害。现有的稻叶瘟发病信息主要通过田间调查来获取，该方法不仅费时费力，而且存在以点代面的代表性差等弊端，难以满足大范围稻瘟病高时效高精度监测的需求。国内外研究多集中在肉眼可见病斑出现之后的病情等级或者严重度估算，很少关注稻叶瘟侵染早期的光谱探测研究，并且缺乏对稻叶瘟高度敏感、可适用于叶片（个体）和冠层尺度（群体）的光谱指数。本项目针对这些问题，借助于先进的光谱分析和空间统计方法主要开展了以下研究：（1）基于近距离成像光谱的稻叶瘟潜在病斑识别；（2）构建基于近距离成像光谱的穗腐病病情严重度估算模型；（3）解析从单叶到冠层尺度稻叶瘟侵染引起的水稻光谱响应规律；（4）构建基于机器学习和特征选择的稻叶瘟早期监测模型；（5）创建适用于多尺度的稻叶瘟敏感光谱指数RIBI，并对多试验点冠层稻叶瘟病情指数进行估算；（6）构建基于RIBI影像热点分析的小农户田块稻叶瘟时空传播动态监测技术。.本研究基于多年温室和野外实验，分别从叶片、近地面冠层和卫星平台获取的多年多试验点实测数据，得到一系列重要成果如下：（1）改进了基于近距离叶片高光谱影像的辐射传输模型反演方法，解析了叶内病斑光谱特性的时空变异规律；（2）提出了穗腐病敏感光谱指数，构建了基于近距离成像光谱数据的穗腐病病情严重度监测方法；（3）构建了一套耦合机器学习与敏感光谱特征选择的分类方法，首次实现了稻叶瘟无症状期的感病叶片高精度识别；（4）创建了稻叶瘟敏感植被指数，显著提升了单叶感病叶片识别和冠层尺度稻叶瘟病情指数估算精度，并基于卫星影像植被指数热点分析，首次揭示了小农户田块稻叶瘟时空演变规律。本研究为基于近距离高光谱遥感的稻叶瘟智能监测系统奠定了基础，为多尺度稻瘟病高光谱监测提供重要理论支撑。项目执行期间发表SCI论文11篇（含RSE论文 4篇），申请发明专利1项，培养硕士与博士研究生5人。项目负责人获中国作物学会青年科技奖，入选农业农村部神农青年英才。

内容获取失败，请点击重试