导　　师： 童庆禧;柳钦火

学科专业： G0503

授予学位： 博士

作　　者： ;

机构地区： 中国科学院遥感应用研究所

摘　　要： 作物类型识别和播种面积估算一直是农业遥感的重要内容，对农作物播种面积和种植结构调整具有重要的现实意义。单一作物像元(纯像元)识别是作物类型识别和作物播种面积遥感估算的关键。 本文设计了野外试验方案，开展了玉米和小麦两个作物全生长期光谱测试实验，获取了大量作物生长期的组分-冠层-像元级光谱、结构参数、背景光谱信息和相关参数，建立了典型农作物波谱库。 基于该典型农作物波谱库，论文对作物纯像元识别和作物面积估算等方面进行了理论与方法的探讨： (1)作物纯像元的识别方法与模型根据作物光谱信息不确定性的原理，通过开发遥感影象-参考波谱距离阈值模型，提出基于距离阈值的纯作物像元的识别方法。对玉米生长后期tm影象试验表明，该方法获得的作物纯像元估算精度可以达到92％；在此基础上，论文提出了结合遥感影象-参考波谱距离阈值、光谱角度和多时相方法对玉米作物纯像元综合识别方法，该方法使玉米作物纯像元的识别精度提高到95％以上。 (2)作物面积估算模型在光谱混合分析模型基础上，提出了光谱角度和影象拟合残差相结合的最优端元选择方法，获得混合像元中各端元的面积比例。通过实地制图试验表明小麦像元内小麦作物比例制图的精度达到95％以上，研究发现3月下旬是小麦亚像元比例面积制图中遥感影象时相的一个较好选择。该方法还对华南一个镇的荔枝种植面积进行了应用，结果表面，荔枝面积估算结果和制图精度达到98％。 (3)像元纯度的检测方法与模型论文分析发现像元纯度指数(ppi)方法在提取纯像元时对端元选择存在不确定性，可能由此导致所提取地物端元纯度降低，或把同类稍有光谱偏差地物作为不同端元对待，使提取的真实地物端元数减少。这种� The crop identification and area estimation are an important domain in agricultural remotely sensing at all times. However, LANDS AT TM images are one of the most important remotely sensing information sources for development of agricultural information extraction methods in a long period. At present, the accuracy, efficiency and cost are still main problems facing the planting area monitoring by wideband TM satellite remote sensing in China. The pure crop pixel identification is of special significance in remotely sensing scaling transformation and inversion of crop biophysical parameters while the estimation of planting area by remotely sensing has great practical meaning for state agriculture structure adjusting at present stage. The convenient and quick pure crop pixel identification and area estimation approach may improve the acquirement ability of agriculture information. In this study spectral library based image-reference spectrum distance /(IRSD/) combined with spectral angle mapper/(SAM/), multi-temporal satellite data/(MSD/) were examined to address the need for pure, spatially accurate maize crop pixel identification. As reference spectra of maize are from in-situ measurement, which less were influenced by other elements such as atmosphere and man's eye in training sample area election etc, they better present the crop average eigenvector in N dimension space than the statistical parameter of training sample area by traditional classification method. Maize crop pixels where IRSD of 6 sites in experimental region were combined were identified with an overall accuracy level of more than 92/%, compared to 51.3/% explained by 6 site training sample method of traditional Maximum Likelihood classification alone. Further, the classification accuracy can attain more than 95/% if SAM and MSD were combined into the IRSD method. This result of this study support the use of moderate resolution spectral imagery combined with TM images to image scaling and crop biophysical parameter extraction. Also, the SAM combi

关 键 词： 纯像元识别 面积估算 农业遥感 农作物 播种面积 种植结构调整 像元纯度

领　　域： [农业科学]