【作者】 刘伟；

【导师】 冯伍法；

【作者基本信息】 解放军信息工程大学 ， 摄影测量与遥感， 2008， 硕士

【摘要】 与全色、多光谱遥感相比,高光谱遥感最大的优势在于能够对地表覆盖类型进行精细探测。成像光谱仪获取的影像光谱分辨率高,可达10nm:波段众多,能为每个像元提供一条完整且连续的光谱曲线。借助从高光谱影像上反演的光谱曲线,通过与标准的参考光谱进行匹配比较,可以直接识别地物目标属性。论文在总结分析高光谱遥感数据预处理技术和光谱特征增强与定量分析方法的基础上,从相似性测度改进和匹配策略优化两个方面对光谱匹配算法进行了深入的研究,主要取得了以下成果:1.以地物光谱特征的匹配识别为目标,总结了光谱数据定标、辐射校正和反射率转换等预处理方法,结合地形要素分类体系分析了典型地物的光谱特征;从基于光谱曲线属性探测的角度出发,研究了光谱曲线特征增强和定量参数提取的主要方法。2.在对高光谱影像模式识别分类方法和光谱匹配技术进行归纳总结的基础上,通过对现有光谱相似性测度的分析和试验比较,提出了一种基于曲线信息熵的光谱相似性测度改进方案。试验表明,与传统的单一测度和简单的综合距离相比,该方法用于光谱匹配能够取得更高的精度和更好的适应性。3.将尺度空间理论引入高光谱影像分类过程,通过对地物光谱信息进行多尺度观察,提取特定尺度下的光谱曲线特征,结合光谱曲线的整体相关性进行匹配分类。试验结果表明:该算法能减少传统匹配方法由于噪声、成像环境等因素引起的错分,有助于分类识别精度的提高。4.结合决策树分层匹配的思想,在对地物反射光谱特征进行具体分析的基础上,根据成像区域地物类型的具体特点,构造了层次分析光谱匹配模型。试验结果表明,该方法通过在不同节点处灵活采用不同的特征参量和匹配策略,能够明显提高目标提取的精度和可靠性。更多还原

【Abstract】 Compared to panchromatic and multi-spectral remote sensing, the greatest advantage of hyperspectral mode rests with its ability of fine detection to the earth’s surface. Hyperspectral images have spectral resolution high to 10nm and a great deal of bands, so it can offer a full and continues spectral curve to any pixel. Therefore, with spectral feature coming from hyperspectral images and standard referenced spectrum, we can make use of the technology of spectrum matching to identify the property of covers on the ground directly. On the basis of summarizing and analyzing the methods of hyperspectral data preprocessing, feature boosting and quantitative analysis, this paper progressed the study in depth on two aspect of improving the comparability measure and matching strategy. The accomplishments include:1. Aiming at matching and recognition to the spectral feature, summarized the spectral data preprocessing methods such as calibration, radiometric correction and reflectivity transformation, analyzed the typical spectral feature combined with topographic feature classification system, and studied the primary means of feature boosting and quantitative parameter extraction to spectrum from the point of the property detection with spectral curves.2. On the basis of summarizing classical pattern recognitive classification methods and spectrum matching technologies , brought forwards a improving programme to the comparability measure through analysis and comparison with experiments. The result indicated that this programme could acquire higher precision and better adaptability in spectrum matching compared with tranditional single measures and simple integrated distances.3. Brought the theory of scale space into the classification of hyperspectral images, extracted the spectral feature in special scales and carried on the matching and classification combined with total relativity between spectral curves. The result made it clear that the arithmetic could reduce the error from imaging noise and environment in traditional methods, and help to improve the precision.4. On the basis of particularly analyzing the spectral reflection and combined with the thinking of matching on different grades from decision tree, constructed a hierarchic structure for spectrum matching according to the character of items in imaging region. The result indicated that this arithmetic could distinctly advance the precision of target extraction, through smartly adopting different feature parameters and matching strategy at different nodes.更多还原