基于近地成像高光谱系统解析小麦白粉病的多尺度和多角度特征

Powdery mildew commonly infest wheat leaves from lower layer to top layer of wheat plants. At present, there are primarily two methods to detect wheat dieases: manual field observations and remote sensing based measurements. However, traditional diagnoses of wheat diseases are usually time-consuming, subjective and inefficient, while conventional studies are being performed by acquiring canopy information in vertical direction based on spaceborne or airborne remotely sensed data. In addition, the influences on middle and lower layers of wheat plants caused by diseases are not fully considered. Consequently, considering the above unsolved questions, ground-based hyperspectral imaging techniques were tentatively utilized in quantitatively identifying wheat powdery mildew in our study. Own-made pushbroom imaging spectrometer (PIS) combining image with spectrum was used to acquire multi-scale and multi-angular hyperspectral images at leaf and canopy sacles. Three studies were primarily performed in this study. 1) According to colour, texture, shape of wheat infested with powdery mildew, the image and specctral characteristics were identified by separating wheat and disease spots from soil background and shadows. 2) Subsequently, multi-scale diagnosis model was built to demonstrate the disease properties of powdery mildew infested wheat by setting the PIS sensor at regular heights from wheat canopies and combining with high resolution IKONOS remotely sensed data. 3) Finally, the PIS sensor was aboard on the multi-angular observation device to obtain disease characteristics of vertical canopies and corresponding middle and lower layers, and then the multi-angular diagnosis model was built. This study can provide ground truth data and lay a methodological foundation for quantitatively monitoring large-scale wheat diseases using spaceborne or airborne remote sensing.

白粉病具有从小麦植株下层叶片逐渐往上层侵染的特点，而传统的病害监测方法以目测手查为主，费时费力、主观性强，航空/航天遥感往往从垂直冠层角度获取病害小麦顶部的信息，没能考虑病害对冠层中、下部的影响。本研究利用近地扫描成像光谱仪PIS，获取高分辨率"图谱合一"的"多角度"和"多尺度"遥感数据，定量解析小麦白粉病单叶和冠层尺度的病害特征。研究内容包括：1）依据染病小麦叶片和冠层呈现的色彩、纹理、形状等病害特征，分离小麦、病斑、土壤和阴影区域，提取纯病害小麦的"图谱"特征；2）设定PIS距离小麦冠层不同高度，联立准同步获取的高分辨率IKONOS航天影像，建立多空间尺度下病害聚类特征和表现规律的"多尺度"诊断模型；3）利用多角度观测架装载PIS传感器，多角度解析染病小麦冠层及中、下层植株和叶片的病害特征，建立小麦白粉病的"多角度"诊断模型。本研究成果将为航空/航天遥感大范围监测作物病害提供方法支撑。

遥感技术为快速、动态、无损、大范围监测和诊断作物病害提供了先进数据源和监测方法。本项目以小麦作为研究对象，基于多源遥感数据解析小麦白粉病和条锈病的“图谱”特征。选取常发型小麦白粉病和条锈病发生、发展和影响产量形成的关键生育期，利用地物光谱仪ASD、扫描成像光谱仪PIS、Image-λ-V10-LU和机载高光谱成像仪PHI，获取病害和正常小麦样地/样本的“多源”高光谱遥感数据，从叶片和冠层两个尺度，解析病害侵染的特点及对应的“图谱”特征。. （1）通过共生矩阵计算影像的纹理变量，得出Data range、Mean和Variance三个变量作为R、G、B波段合成的假彩色影像能更好地突出叶片上的白粉病斑。之后，选取对白粉病敏感的10个红波段（675.1-681.1 nm）和10个近红外波段（706.2-712.1 nm）进行平均后构建特征空间，得到提取病斑的线性方程（y=3.48*x-7.57），总体分类精度为92.5%。结果表明，白粉病斑的存在极大增强了影像的纹理特征。. （2）在小麦灌浆期，对比分析ASD探头垂直冠层（0°）、与冠层呈45°和90°三个方向的高光谱反射率。结果表明，随着观测角度的变化，视场中的植被和土壤比例会变化，轻度和重度感染白粉病的小麦冠层在三个观察角度的反射率表现出相似的规律，即90°测量下的反射率最高，轻度小麦冠层最高超过85%，重度的小麦冠层也超过了60%，与其它两个角度的值相差最大；0°测量下的反射率次之；45°测量下的反射率最小，且与0°条件下的值相差较小。. （3）利用ASD叶片夹测取了拔节期和灌浆期各110片样本的正反面光谱反射率。基于光化学反射率指数PRI建立了反演模型。结果表明，叶片正面的反射率小于背面。拔节期得到的反演模型的决定系数R2为0.74，而灌浆期的R2为0.83。灌浆期比拔节期更适合评价小麦条锈病的侵害等级。冠层尺度研究中，分析三个生育期正常和病害冠层的PHI影像光谱和影像特征，建立病情指数与红（621.0~720.1 nm）和近红外（768.7~805.7 nm）波段平均值的二元线性方程，得到相关系数为0.923，标准差为0.108。结果表明，从拔节期到乳熟期病害严重度不断增加。. 本研究得到的部分解析方法和反演模型，可为基于航空和航天遥感数据监测大尺度区域的作物病害提供方法参考。