多源数据小麦病害遥感识别与监测方法研究

The yield loss of crops due to crop diseases and pests are increasing under the context of global climate change. For monitoring of crop diseases and pests, the traditional methods that are adopted by major plant protection department of our country are still through manually inspection and field sampling. However, these methods are not only time consuming, labor intensive, but are also unable to obtain information about disease occurrence and dispersal over vast areas. Nowadays, incorporating datasets from multi-sources for disease monitoring and forecasting, such as remote sensing data, agricultural data as well as meteorological data, is becoming an important tendency. The present proposal aims at solving a series of problems in crop diseases monitoring and forecasting. These problems include: how to separate crop diseases from other stressors？how to utilize satellite images for monitoring? and how to forecast crop diseases at a finer spatial scale? Two important diseases of wheat, powdery mildew and yellow rust, are adopted as the objects of our study. A control experiment is planned to simulate wheat plants under different stresses (e.g. diseases, nutrient stress and drought). The divergence between wheat diseases and other stresses can be analyzed at different dimensions (e.g. spectral, temporal, etc.). Based on which, a knowledge base will be established. The methodologies for remotely sensed identification of disease, and for disease severity determination will also be developed. In monitoring diseases through satellite images, we will study the techniques for extracting temporal and spatial features from some commonly used multi-spectral image sequences. Then, based on the mixture tuned matched filtering technique, we will develop a disease identifying and monitoring framework which will integrate spectral, spatial, temporal and other types of information from both moderate resolution and high resolution images. Moreover, a forecasting model of wheat diseases will be established that is able to incorporate meteorological, agricultural and remote sensing datasets. The outcomes of this proposal will be of great importance for ensuring food security, guiding fungicide spray and minimizing the contamination caused by overuse of fungicides.

全球变化引起病虫害对农业生产造成的损失日益加剧，目前我国农业植保部门对病害的监测方法主要是目测手查法，这种传统的以"点"代"面"的工作模式费时、费力，且无法获取大面积的病害发生程度和蔓延信息。综合遥感、植保和气象多源数据的作物病害监测和预报成为重要趋势。项目针对目前病害遥感监测中病害与其它胁迫类型易混淆，基于卫星的病害监测方法欠缺，病害预测主要基于气象数据预测空间尺度过大等突出问题，以小麦主要病害白粉病和条锈病为对象，设置控制实验，研究小麦病害和营养、干旱等胁迫在光谱、时序等多维特征上的差异，构建小麦病害特征知识库，发展病害遥感识别和严重度监测方法；研究基于混合调谐匹配滤波分析的融合中、高分辨率卫星影像光谱、空间、时间等多维特征的大范围病害识别监测方法；构建综合气象、农学、植保、遥感数据的小麦病害预测预报模型。对保障我国粮食安全和指导适时、合理施药保护环境具有重要意义。

全球变化引起的病虫害对农业生产造成的损失日益加剧，目前我国植保部门对病虫害的监测及预测方法主要是目测手查法，这种传统的以“点”代“面”的工作模式费时、费力，且无法及时获取大面积的病害发生程度和蔓延信息。本项目围绕作物病虫害胁迫引起的叶片光谱吸收特征的变化信息，从以下几个方面对小麦主要病虫害的遥感监测及预测方法进行了研究：1）针对冬小麦主要病虫害胁迫下的叶片首先表现的生理生化组分的变化，利用高光谱数据，对不同病虫害种类引起的光吸收特征的变化进行定量提取，建立了不同作物病虫害的识别模型6个，提出不同病害胁迫的植被指数4个，敏感特征类型3种；2）基于田间实测个冠层高光谱反射率，利用小波变换在信号处理与特征提取方面的优势，对不同病害的光谱特征进行了分解和重构，提取了敏感的小波特征，并构建了冠层尺度的小麦病害的识别和严重度反演方法。3）在此基础上，综合区域尺度上的遥感、植保、气象等多源数据，利用支持向量机、相关向量机、神经网络等智能算法，实现了冬小麦主要病虫害的胁迫信息从分散的地面“点状”观测向区域尺度的“面状”栅格信息拓展,并对作物病虫害发生发展的生境信息与作物生长模型相结合的时空耦合和同化机理研究，建立作物主要病虫害的遥感监测和预测预报模型，依次开展基于生境信息的作物病虫害时空发展规律的监测预警机理研究。4) 率先建立了全国尺度业务化运行的“作物病虫害遥感监测与预测系统”，定期发布了26期全国范围小麦、水稻、玉米等主要作物病虫害预测预报报告，被农业部全国农业技术推广中心权威发布全国作物病虫害发生状况的“植物病虫情报”和全球公认的三大农情遥感监测系统之一的“全球农情遥感速报系统”（CropWatch）采纳。本项目的研究成果为作物病虫害的精准防空、变量施药等提供了科学的理论依据和指导。