【作者】 杜世州；

【导师】 王荣富；

【作者基本信息】 安徽农业大学 ， 作物栽培学与耕作学， 2013， 博士

【摘要】 小麦白粉病是我国小麦生产中主要的病害之一,病害发生时,单叶尺度和冠层尺度均表现出一定的特征信息,能够通过高光谱特征表征和解译,适合于采用遥感方式进行的监测技术研究。本研究于2011年、2012年利用地面高光谱遥感和HJ-CCD影像对小麦田间白粉病的生理生化指标和病情严重度进行了定性或定量的反演评价。(1)研究了小麦白粉病叶片和植株生理生化特征与病情严重度间的相关关系。小麦白粉病病情严重度与叶绿素含量、叶绿素密度间的差异达到显著或极显著水平。叶绿素含量与病情严重度间存在极显著的负相关关系,R2达到0.8466以上；叶绿素密度与病情指数间存在极显著的负相关关系,R2达到0.9725以上；叶片全氮含量与病情指数间达到极显著负相关关系(R2=0.8655)。叶片病情严重度与SPAD值呈显著负相关,其决定系数达到0.9257；叶片病情严重度与Ch1、NBI指标呈显著负相关(RCh1=-0.768*,RNBI=-0.632*),叶片病情严重度与Flav呈显著正相关(RFlav=0.727*)。通过高光谱技术对NBI指数进行反演和校正,LS-SVM建模得到的R和RMSEP分别为0.9475和1.0037,优于BPNN所建模型的预测效果。尝试利用叶绿素成像荧光系统检测小麦白粉病叶片的生理状态和快速光响应曲线的变化规律,比较两种选择模式条件下,荧光成像特征的变化规律,比较了小麦病害叶片与健康叶片荧光成像的异质性,表明了受病害胁迫叶片,尤其是表面存在病斑的叶片,荧光参数间存在较高异质性。提出对光曲线拟合方程进行一阶求导得到曲线斜率方程,来表征快速光曲线的动态变化,明确基于荧光成像系统检测小麦白粉病叶片的合理性。(2)分析了叶片光谱敏感度和原始光谱对不同病情等级叶片尺度的响应关系,明确了630-680nm波段作为小麦白粉病病害等级响应的最佳敏感波段。利用植被指数法、连续统去除法和小波变换处理来反演单叶尺度小麦白粉病病情严重度。初步筛选确定了16个基于高光谱数据构建的植被指数。除了RVSI和PSRI指数外,其他植被指数与病情严重度间相关系数呈显著性差异,其中指数GNDVI、NDVI、SR、PRI、 OSAVI、SAVI、SIPI、VARI、IWB、RVSI和PSRI与病情严重度呈负相关关系,其他植被指数与病情严重度呈正相关关系。以AAI为自变量建立回归模型,取得了更好的反演效果(R2=0.9195,F=251.3122)。基于db6小波系对病叶原始光谱进行降维处理,筛选出了7个通过0.01水平检验的小波能量系数,作为小麦白粉病严重度的识别监测处理。归纳总结了共48个光谱特征与病情严重度(DI)的相关分析结果并对估算模型进行精度检验。其中,41个特征参数均与DI有显著的相关性达到极显著。(3)研究了冠层尺度小麦白粉病不同病情指数光谱响应特征,通过植被指数和病情指数相关性分析,筛选出小麦白粉病病情指数反演监测模型。选取测试的25个植被指数光谱特征中,除了冠层尺度植被指数PSRI和WI与病情指数间相关性不显著外,其余23个冠层尺度植被指数光谱特征均与小麦白粉病病情指数间显著相关。基于选用基于负熵为信号统计量的FastICA算法对高光谱数据进行降维分析,基于PLS提出多特征变量的小麦白粉病病情指数监测模型,并对可见光、近红外波段的数据进行分离,构建小麦冠层病害胁迫特征三维空间分布模型。经复相关系数和均方根误差判别,RIC1-NIRICl-λrep模拟方程具有较高的复相关系数和较低的均方根误差,其模型预测值具有较好的精度。(4)通过变化向量分析,对冬小麦白粉病害灾情进行遥感监测,利用冬小麦染病面积与绿波段归一化植被指数的响应关系,重新定义了矢量空间,构建了冬小麦白粉病的遥感监测模型。采用多种植被指数建立空间变化向量分析,通过分析6种植被指数的变化范围,得出了绿波段归一化植被指数对小麦白粉病更为敏感,该指数可用来建立小麦白粉病的监测模型。通过分析3个地区的病害遥感监测图,得出晋州北部地区病害较严重,宁晋县中部有中度的病害,而藁城市北部和中部则有大面积的病害发生,应根据3个地区不同的病害程度,进行喷药控制,以降低病害带来的损失。更多还原

【Abstract】 Wheat powdery mildew is one of the major diseases in wheat production in China. The leaf and canopy scale were both showed some characteristic information when this disease occurred. As such information can be interpreted by the hyperspectral feature, this disease is very suitable for monitoring by using remote sensing. In this paper, the physiological and biochemical parameters of wheat powdery mildew had been qualitative or quantitative inverted by hyperspectral data and HJ-CCD image, which were acquired in2011and2012.(1) The correlation between physiological and biochemical characteristics and disease severity of wheat powdery mildew had been analyzed. The results showed that wheat powdery mildew disease severity and chlorophyll concentration had a significantly negative correlation (R2>0.8466), meanwhile the highly negative correlation (R2>0.9725) had also been found between disease severity and chlorophyll density. In addition, the disease severity and leaf total nitrogen, SPAD, Chi, NBI were showed negative correlation, the coefficient of determination were0.8655,0.9257,0.768and0.632, respectively. Finally, the disease severity and Flav had a significant positive correlation (RFiav=0.727*). Through the inversion and calibration for NBI index by hyperspectral technology, the R and RMSEP of LS-SVM had reached0.9475and1.0037, which is better than the BPNN.The variation between physiological state and fast light response curve had been measured by using chlorophyll fluorescence imaging system. The fluorescence imaging feature, heterogeneity between disease and health leaves had been compared under two imaging modes. The results showed that the fluorescence parameters had high heterogeneity when the leaves had been infected, especially some disease spot leaves. The dynamic change of light response curves had been obtained by the slope equation of curves which had been fitted by first-order derivative. These results indicated the rationality of the detection of powdery mildew of wheat leaf based on fluorescence imaging system.(2) The response between hyperspectral reflectance and different disease severity had been analyzed, the best sensitive waveband (630-680nm) had also been identified for detecting the wheat powdery mildew. The disease severity based on the leave level was inverted by using vegetation index, the continuum transformation and wavelet transformation method. Sixteen hyperspectral vegetation index had been used to invert the disease severity, the correlation coefficient all passed the significance test (P-value<0.05) except RVSI. PSRI. GNDVI、NDVI、SR、PRI、OSAVI、SAVI、SIPI、VARI、IWB、 RVSI and PSRI had showed negative correlation, whereas the other indexes had positive correlation. The regression model built by AAI has a better inversion result (R2=0.9195, F=0.9195). The original spectrum dimension had been reduced by db6wavelet. Seven wavelet energy coefficients passed the significance test (P-value<0.01). The DI and48spectral features had been analyzed, and the model precision had also been evaluated. The results show that41spectral feature parameters has significantly correlation with DI.(3) The spectral response between disease severity and hyperspectral reflectance based on the canopy scale had also been analyzed, the inversion model of wheat powdery mildew had been built through the correlation analysis between vegetation index and the disease index.25vegetation indexes which were selected had showed significant correlation with the DI, except the PSRI and WI. The reduction of dimension of hyperspectral data were based on the FastICA algorithm, then the monitoring model of wheat powdery mildew disease index was built by PLS. This model can separate the data in visible and near infrared waveband, thus another three dimensional space distribution model of disease stress characteristics can be built. RIC1-NIRIC1-λrep simulation equation had a high correlation coefficient and low mean square error through the discrimination of these two statistical parameters, and a good accuracy can be obtained by this equation.(4) The remote sensing monitoring model of wheat powdery mildew had been built by the change vector analysis, the method used the response relationship between wheat infected area and GNDVI, and redefined the vector space. Several vegetation indexes had been employed for the change vector analysis, the GNDVI was more sensitive to the wheat powdery mildew disease by calculating the variation range of6vegetation indexes. By means of analyzing the disease of three region, the northern region of JinZhou city had a severe disease, the central region of NingJin county had moderate disease, and the large area disease had been found in the central of GaoCheng city. These results showed that the pesticide should be applied in these area for lowing the the loss of wheat.更多还原