节点文献
水稻主要病虫害胁迫遥感监测研究
Monitoring the Rice Disease and Insect Stress with Remote Sensing
【作者】 刘占宇;
【导师】 黄敬峰;
【作者基本信息】 浙江大学 , 农业资源利用, 2008, 博士
【摘要】 本研究通过对浙江省和黑龙江省6个县(市、区)5种水稻病虫害的观测,运用多种数据处理方法,选择对水稻病虫害响应的敏感光谱区域和谱段,构建病虫害胁迫指数,探索对水稻不同病虫害的危害等级分类和色素含量、病害严重度指数、虫情指数等危害指标的估算方法研究,运用QuickBird影像提取稻飞虱危害面积和产量损失评估。研究内容和研究结果,概述如下:(1)受病虫害胁迫水稻的光谱特征变化分析除稻飞虱和穗颈瘟引起的倒伏外,水稻植株受病虫害胁迫后,光谱反射率在可见光范围内上升,在近红外和短波红外区域内下降。水稻在受到病虫害胁迫后,“红边”和“蓝边”均发生蓝移,向短波方向偏移10nm左右;“绿峰”和“红谷”则均发生红移,向长波方向偏移8nm左右。(2)水稻病虫害响应敏感的光谱波段选择及病虫害胁迫光谱指数的构建运用连续统去除法、光谱敏感度分析法和农学参数-光谱相关系数法等三种方法,对健康的和受病虫害胁迫水稻的原始光谱、反对数光谱、一阶、二阶微分光谱进行分析,从各种变换形式的光谱数据中筛选出对病虫害响应敏感的光谱区域为460-520 nm、530-590 nm、620-680 nm和红690-730 nm,并构建了22个水稻病虫害胁迫光谱指数.(3)水稻不同病虫害危害等级的识别方法研究运用聚类分析CA、概率神经网络PNN、学习矢量量化LVQ神经网络和支持向量分类机SVC(C-SVC)等四种分类方法,开展对水稻不同病虫害的危害等级识别研究,并从分类精度、使用方便程度和消耗时间三个方面来评判分类方法的优劣。其中,分类精度排序为PNN网络(93.5%)>C-支持向量分类机(90.5%)>聚类分析CA(84.3%)>LVQ网络(83.2%),使用方便程度来排序为PNN网络>C-SVC>LVQ网络>聚类分析CA,消耗时间排序为C-SVC<PNN<LVQ<CA。(4)色素含量和病害严重度指数的估算方法研究运用多元逐步回归分析MSR、偏最小二乘回归法PLS、径向基函数RBF神经网络、后向传播BP神经网络、支持向量回归机SVR等五种回归模型,以及现有的14个光谱植被指数、17个高光谱三边特征参数和本研究提出的22个病虫害胁迫指数的简单线性回归、二次多项式回归模型等多种估算方法,对受病虫危害的叶片色素含量和稻胡麻斑病病害严重度指数进行估算方法研究,并使用相关系数R、方差分析F检验值、均方根误差RMSE、平均绝对误差MAE和平均相对误差MRE等五个指标,对不同的估算方法进行综合评价.(5)基于QuickBird影像的稻飞虱危害评估研究通过目视解译提取研究区内的土地利用类型专题图,选择水稻种植典型样区.研究发现:倒伏的、收割的和受稻飞虱危害而尚未倒伏的稻田与健康稻田相比,其影像色调及NDVI、EVI均发生了巨大的变化。对典型样区内受稻飞虱不同危害状况的稻田进行矢量化,结合实地调查的产量损失数据,确定2005年夏秋之际爆发的稻飞虱,对当年晚稻的产量造成的损失比例为40%。
【Abstract】 The specific objectives of this research were threefold. The first objective was to select the spectral region and wavebands, which were sensitive to rice disease and insect stress, and to develop the stressed spectral indices with the multiform data processing methods through five species of rice disease and insect in Zhejiang Province and Heilongjiang Province. The second objective was to explore the discrimination of different endangered categories of rice disease and insect and estimation of harm indices such as pigment content and disease severity index. The third was to extract the endangered paddy area caused by rice planthopper and assess the yield loss from the QuickBird images. The summary study content and result were presented as follows:(1) Change analysis of hytperspectral characterization caused by rice disease and insect stressWhen disease and insect stressed rice, with the exception of lodged rice caused by rice planthopper and rice panicle blast, the hyperspectrl reflectance increased in the visible region and the hyperspectrl reflectance decreased in the near infrared and shortwave infrared regions. The red edge and blue edge also shifted toward the short spectral region about 10 nanometers. The green peak and red trough also shifted toward the longer waveband 8 nanometers.(2) Selection of sensitive spectral region and development of stressed spectral indices responding to rice disease and insect stressThe raw spectra, inverse logarithmatic spectra, first-order and second-order derivative spectra of healthy and non-healthy rice leaves caused by disease and insect was analyzed with three analysis methods, which are the continuum removal, spectral sensitivity analysis and correlation coefficients between agricultural parameters and spectral reflectance. The sensitive spectral regions, which were more sensitive to rice disease and insect, were located in the blue region (460 -520 nm), green region (530-590 nm), red region (620 - 680nm) and the red edge region (690-730 nm) in despite of different transformation spectral reflectance. And then 22 stress indices of disease and insect were developed in this study.(3) Study of discrimination method on the endangered categories of different rice disease and insect stressCluster analysis (CA), probabilistic neural network (PNN), learning vector quantization neural network (LVQ) and C-support vector classification (C-SVC) machine were utilized to discriminate the endangered categories of different rice disease and insect stress consisting of rice brown spot, rice Aphelenchoides besseyi Christie, rice panicle blast, rice planthopper and rice leaf roller. Three indices such as classification accuracy, convenient utilization degree and consuming time were applied to assess the above-mentioned four discrimination methods. The successive order of classification accuracy were PNN (93.5%),C-SVC (90.5%),CA (84.3%) and LVQ (83.2%). The successive order of convenient utilizations degree were PNN>C-SVC>LVQ>CA, the anterior the position, the higher the convenient utilization degree. The successive order of consuming time were C-SVC<PNN<LVQ<CA, the posterior the position, the longer the consuming time.(4) Study of estimation method on the foliar pigment content and disease severity index (DSI)Three kinds of regression methods were utilized to estimate the foliar pigment (i.e. total chlorophyll) content of rice Aphelenchoides besseyi Christie and disease severity index (DSI) of rice brown spot. The one kind was data mining techniques consisting of radial basis function (RBF) neural network, back-propagation (BP) neural network and support vector regression (SVR) machine. The second was multiple regression analysis including the multiple stepwise regressions (MSR) and partial least squares (PLS) regression. The third was simple linear regression and binomial regression equation, which were composed by the 14 existing spectral vegetation indices, 17 hyperspectral feature parameters and 22 stress indices of disease and insect developed in this study. The correlation coefficient (R), analyze of variance (F test), root mean square of prediction error (RMSE), mean absolute error (MAE) and mean relative error (MRE) were applied to assess comprehensively the above mentioned three kind of regression methods.(5) Study on yield loss assessment due to rice planthopper based on QuickBird imagesThe thematic map of land use was extracted from the fused wavelet image by visual interpretation, and the sample rice planting area was chosen from the thematic map of land use. The image tones, normalized difference vegetation index (NDVI) an ratio vegetation index (RVI) of lodged rice, harvested paddy field and potential endangered rice changed heavily compared with healthy rice. The yield loss caused by rice planthopper in 2005yr accounted for 40% of the total rice production through quantifying the different endangered categories of paddy field and combining with the data of rice yield loss from one farmer to the others.
【Key words】 Disease and Insect Stresses; Spectral Vegetation Index; Sensitive Spectral Wavebands; Neural Network; Rice; Remote Sensing; Support Vector Machine;