基于激光点云和视觉计算的阔叶树林学参数反演研究

This topic take terrestrial laser scanner (TLS) as the means for physiological information measurement of broad-leaved trees, also combine with forest science and computer vision theory to calculate accurate forestry parameters of broad-leaved trees from point cloud data of laser scanner. Specific research plan include: Firstly, three-dimensional point cloud feature extraction and classification algorithms are designed for classification and identification of broad-leaved trees organs; Secondly, combing the computer graphics theory, algorithms are proposed for mapping discrete point cloud to the 3D surface of real trees, thus the real morphological models of broad-leaved trees are reconstructed in order to calculate tree parameters include stem form exponents, branches and knots positions; Thirdly, based on the botany and physics principle, realistic rendering methods of broad-leaved tree leaves under different ecological conditions and natural environment are realized, which aim to provide scientific evidence for leaf area index and canopy structure parameters estimate; Lastly, compared with the forestry index that acquired by leaf area measurement Instrument, hyper spectral radiometer and Airborne Lidar, validity of our method are verified and error correction are considered. In brief, this topic use the latest measurement technology to extend traditional tree index acquisition method, and combine multidisciplinary theory to seek biomass values from TLS point cloud data. After this subject study, the modes of broad-leaved trees measurement are broaden and the construction of Chinese forestry digitization process is developed.

本课题运用地面激光扫描仪作为阔叶树信息计测的手段，并结合林学与计算机视觉的理论方法从激光点云数据中挖掘精确的阔叶树林学指标。具体的研究思路为：首先，设计三维点云的特征提取与分类算法，实现阔叶树不同植物器官的识别与分类；其次，结合计算机图形学理论研究离散点云到三维曲面的映射算法，重构阔叶树真实的三维形态模型，进而获取阔叶树树干的干形指数、节瘤、分枝位置等参数；再次，基于植物学与物理学原理，实现阔叶树叶片在不同生态环境下的真实感绘制，旨在为叶面积指数和树木冠层结构参数的测量提供依据；最后，与叶面积测量仪、高光谱辐射仪、机载激光雷达等其它途径获取的阔叶树林学指标进行类比，验证结果的有效性并修正误差。本课题以最新的测量技术丰富了传统的阔叶树数据获取的方法，并借鉴不同学科的理论挖掘蕴含其中的生物量参数，进而为阔叶树林学指标的计测研究和我国"数字化林业"的建设进程拓宽了思路。

本课题运用激光扫描与视觉计算相关算法获取阔叶树林学相关生理参数，做出工作如下：1) 完成了面向点云的去噪与拟合算法，设计了基于点云的正交最小二乘、移动最小二乘与叶边界精确提取方法，消除扫描中叶子在风中抖动产生的噪声点。2) 运用球极投影方法把林木激光点云投影到平面上获取半球图像，并根据beer-lambert定律来计算有效叶面积指数，并与其他设备获取结果相互比较进行结果验证。3) 提出了结合点云的法矢量分布、结构张量、局部切平面等点云特征，用于林木点云的自动分类，进而识别林木中不同的器官，如枝、叶、果实等。4) 提出了一种基于Laplace收缩算法与势能场及距离度量框架下的面向点云的活立木枝干骨架算法，发展了一套完整的基于活立木点云数据的枝干建模算法。5) 提出了一种基于扫描距离与最小偏转角的自适应阈值的三角剖分算法，改算法棵自动实现点云向空间叶面的转换，进而获取植物的真实叶面积值。6) 运用机载激光雷达获取南方林段的激光点云数据，设计了空间分水岭算法与形状拟合的株株分离方法，同时初步实现了单株树体中的叶叶分离算法。本课题开展期间共发表学术论文36篇，软件著作权15余项，申请各类专利6项，为激光点云在林业中的测绘与计量奠定了坚实的理论基础。