【作者】 赵旦；

【导师】 李增元；

【作者基本信息】 中国林业科学研究院 ， 森林经理学， 2012， 博士

【摘要】 随着遥感技术的快速发展，传统光学遥感虽然在林业领域的应用广泛，但仅仅能提供简单的空间信息和光谱信息，远远无法满足森林资源调查的需求。近十年来，激光雷达（Light Detection and Ranging，LiDAR）技术和高光谱遥感技术分别凭借高精度的三维空间信息和高光谱分辨率，迅速在各行各业得到广泛应用。高密度的LiDAR点云数据与高空间分辨率的高光谱数据不仅能够提供林分尺度的森林参数，还能提供单木尺度的森林参数。因此，本文以利用LiDAR数据和高光谱遥感数据获取单木尺度的参数为目的，进行了以下几个方面的研究工作：（1）针对冠层高度模型（Canopy Height Model，CHM）中普遍存在的无效值，提出了冠层控制的概念。冠层控制近似地决定了CHM中真实树冠区域，忽略无效值的存在，这为后续的CHM优化和单木分离提供了较为准确的树冠范围。冠层控制主要使用形态学闭运算进行，树冠区域通过由目标区域枝下高决定的冠层控制阈值在闭操作后的CHM上判断获得。（2）结合冠层控制的CHM优化算法。CHM的质量直接或间接地影响了基于CHM的各种森林参数提取精度，因此对CHM进行优化是十分必要的。无效值的存在是影响CHM质量的主要因素之一，为了只填充无效值而不填充冠层间隙，本文提出结合冠层控制的CHM优化算法，算法首先使用拉普拉斯算子从CHM中寻找可能无效值，然后引入冠层控制以确定其是否落在树冠范围内，是则属于无效值，反之则是冠层间隙，最后采用全局平滑的CHM对应像素对无效值进行填充。（3）结合冠层控制的分水岭郁闭林区单木分离算法。经过优化的CHM已经适合进行单木尺度的森林参数提取，但是郁闭林区的单木分离仍然是LiDAR单木分离的难题之一，因此本文针对郁闭林区提出了结合冠层控制的分水岭郁闭林区单木分离。算法引入冠层控制进行外部标记（树冠区域）的确定且辅助内部标记（树顶）的查找，根据外部和内部标记进行简化标记分水岭变化，最后结合归一化的点云得到单木的位置、树高和冠幅三个参数。（4）基于LiDAR和高光谱遥感数据的单木树种识别算法。获取单木的位置、树高和冠幅之后，结合高空间分辨率的高光谱遥感影像获得单木的树种成为提高后续单木参数估计精度的重要步骤之一。本文基于LiDAR数据的单木分离结果，从高光谱遥感影像中提取单木树冠区域对应的光谱并优化，再对优化的单木光谱分别进行支持向量机（Supporting Vector Machine，SVM）和光谱角制图（Spectrum Angle Mapper，SAM）法分类得到单木树种。通过在黑龙江凉水试验区进行试验的结果和甘肃大野口试验区补充试验的结果表明，冠层控制能够有效地在CHM上恢复真实冠层区域；结合冠层控制的CHM优化填充了CHM上大多数的无效值，同时保留了冠层间隙；结合冠层控制的分水岭郁闭林区单木分离能较好地识别目标区域的优势与亚优势单木，识别部分中庸木，提取的单木树高精度超过90%；基于LiDAR和高光谱遥感数据的单木树种识别能够较好的识别大多数的单木树种，优势树种的识别正确率大于90%，优势与亚优势树种识别平均正确率大于70%。总之，LiDAR和高光谱数据结合能够很好地提供单木尺度的森林参数。通过冠层控制优化CHM，可以更高质量地获得单木的位置、树高和冠幅，配合高光谱数据的光谱信息则可以获得单木树种信息，为提高后续的各种单木或林分森林参数的估计精度奠定了基础，因此，本文的研究工作在森林资源调查中具有巨大的实用价值和应用前景。更多还原

【Abstract】 Following the rapid development of remote sensing techniques, traditional passive optical remote sensing is applied in forestry more and more broadly. However, its relatively simple spatial and spectrum information can’t satisfy the requirement of forest resource investigation. Fortunately, Light Detection and Ranging (LiDAR) and hyper-spectrum techniques are applied in various fields including forestry quickly by the high quality3-dimension spatial information and high spectrum resolution respectively. High density LiDAR point cloud data, as well as the high spatial resolution hyper-spectrum image, can provide not only forest parameters in stand scale, but also in individual tree scale. To obtain individual tree parameters using LiDAR and hyper-spectrum data, following topics were studied:(1) A concept, named crown control, was proposed for the invalid values which were regularly existed in canopy height model (CHM). The crown control determined the real tree crown areas in CHM approximatively, neglected existing of the invalid values. Relatively exact crown areas were provided for the following CHM optimization and individual tree isolation. The crown contrl was carried out using morphological closing operation. The crown areas were determined by a threshold, which was decided by the crown base height of target area, based on the CHM after closing operation.(2) Filling Invalid Values in CHM with Morphological Crown Control was studied. The quality of CHM directly or indirectly influences the accuracy of various forest parameters extraction based on CHM. As a result, it was necessary to optimize CHM. The existing of the invalid values was one of the main factors which influenced the CHM quality. A CHM optimization algorithm with crown control was proposed to fill the invalid values, and retain canopy gaps. The algorithm firstly found potential invalid values in CHM using Laplacian operator. Then these potential invalid values located in canopy area by crown control, they were invalid values, otherwise canopy gaps. Finally the invalid values in CHM were replaced by corresponding pixels in global smoothed CHM. (3) Isolating individual trees in closed forest using watershed with crown control. The optimized CHM was suitable for the forest parameters extraction in individual tree scale. However, isolatiog individual trees in closed forest was a difficult problem. This paper proposed a watershed with crown control method for this issue. The method imported crown control to determine outer markers (crown areas) and assist find inner markers (treetop). Then simplified marker controlled watershed transform was applied to CHM with outer and inner markers. At last, the position, height and crown radius were gained from the watershed results and height normalized point cloud.(4) Individual tree species identification based on LiDAR and hyper-spectrum data. Once the individual tree position, height and crown radius were gained, individual tree species identification based on LiDAR and hyper-spectrum data should be an important step to improve the accuracy of subsequent tree parameters estimations. The method firstly extracted tree spectrum from hyper-spectrum image based on the individual tree isolation results, and then combined them. Then the classification was carried out with the combined individual tree spectrums using Supporting Vector Machine (SVM) and Spectrum Angle Mapper (SAM) respectively. Lastly, the tree species were obtained.Through the experiments in Heilongjiang Liangshui sample plot and the supplement experiment in Gansu Dayekou sample plot, some conclusions were concluded. The crown control recovered real crown areas in CHM effectively. The CHM optimization with crown control filled the most invalid values in CHM, as well as retained canopy gaps. The individual tree isolation with crown control found the most dominant and subdominant trees, and some medium trees in the plot well, and the accuracy of the tree height estimation was greater than90%. The individual trees species identification based on LiDAR and hyper-spectrum data identified dominant tree species in the plot with correctness of over90%, and sub dominant tree species with correctnes of over70%.In summary, LiDAR and hyper-spectrum data provide good forest parameters in individual tree scale. Higher quality individual tree position, height and crown radius can be extracted using optimized CHM with crown control. And the hyper-spectrum data can provide individual tree species, which is useful for the subsequent forest parameters estimations. As a result, these studies have high pratical values and application prospects in forest resource investigations.更多还原