基于机载激光雷达的林木特征研究

发布时间：2018-03-14 02:17

本文选题：激光雷达　切入点：点云数据　出处：《中南林业科技大学》2014年硕士论文　论文类型：学位论文

【摘要】：森林资源作为最大的陆地资源,它的变化不仅关系着社会经济的发展,而且对生态环境也有巨大的影响。近年来人类环保意识的提高,森林资源的重要性得到更加广泛的认知。因此高效地获取森林资源状况成为必然的趋势。森林是由单株的树木构成,准确的森林单株木参数是获得详细森林资源状况的保证,所以准确获取高精度单株木参数意义深远。虽传统的光学遥感在林业领域应用广泛,但是仅仅能够提供简单的空间和光谱信息,远远无法满足森林资源调查的需求。机载激光雷达技术,作为现代对地观测的新技术之一,与其它遥感测绘方法相比,该技术的主要优势为能够快速、直接的获取地物表面的三维坐标数据。因此,近几年来在森林资源调查、城市三维测绘等领域取得了广泛的应用。在森林植被地区,由于机载激光雷达系统发出的激光束有一定的穿透性,可穿透植被到达地表。因此利用此技术,可以获取树冠表面和地表面的三维数据。本文以利用LiDAR数据获取单木尺度的参数为目的,进行了以下的研究工作： (1)树冠高度模型的获取。本文运用不规则三角网滤波算法对雷达点云进行滤波处理,通过滤波分出地面点、植被点、异常点、房屋点等,达到了很好的分类效果,将分类的地面点及植被激光点分别生成数字地面模型和数字地表模型,二者做差运算得到树冠高度模型。 (2)树冠高度模型优化算法。树冠高度模型中包括了很多高程漏洞直接或间接地影响了基于树冠高度模型的各种森林参数提取精度。本文提出了一种新的方法解决这个问题,通过形态学的闭运算获取到平滑的树冠高度模型,再通过树冠高度模型矩阵的归一化、二值化、卷积等运算过程,用平滑高程值替换掉树冠高度模型中异常的高程值点,树冠与树冠之间的低点保留下来,使得连续的树冠被修复和对齐。 (3)对优化后的树冠高度模型进行多尺度单株木分离。面向对象的多尺度分割是从一个像素的对象开始进行一个自下至上的区域合并技术,小的影像对象可以合并到稍大的对象中去。本研究运用面向对象的多尺度分割方法,将树冠高度模型生成图像对象原型,通过设置不同的尺度参数来实现区域生长算法,完成了对研究区域的树冠高度模型的分割,取得了理想的分割效果。 (4)最后通过建立雷达估测林木特征值与实测林木特征值的线性回归关系,对单株木的树高和冠幅进行反演。冠幅估测的平均精度达到88%,树高估测的平均精度高达89%。
[Abstract]:Forest resources as the largest land resources, its changes not only related to the development of social and economic, but also has a huge impact on the ecological environment. The importance of forest resources is more widely recognized. Therefore, efficient access to forest resources has become an inevitable trend. Forests are made up of individual trees, and accurate parameters of individual forest trees are the guarantee of obtaining detailed information on forest resources. So it is of great significance to accurately obtain the parameters of single tree. Although traditional optical remote sensing is widely used in forestry, it can only provide simple spatial and spectral information. The airborne lidar technology, as one of the new technologies of modern earth observation, has the advantage of being able to quickly compare with other remote sensing mapping methods. Therefore, in recent years, it has been widely used in forest resource survey, urban three-dimensional mapping and other fields. Since the laser beam emitted by the airborne lidar system can penetrate the vegetation to the surface of the earth, the 3D data of the crown surface and the ground surface can be obtained by using this technique. In order to obtain the parameters of single tree scale using LiDAR data, the following research work has been done in this paper:. This paper uses irregular triangular network filter algorithm to filter radar point cloud, through filtering out ground points, vegetation points, outliers, housing points and so on, achieves very good classification effect. The classified ground points and the vegetation laser points were used to generate the digital ground model and the digital surface model respectively. The crown height model was obtained by the difference operation between the two models. This paper presents a new method to solve this problem, which includes many height holes that directly or indirectly affect the extraction accuracy of various forest parameters based on tree crown height model. The smooth crown height model is obtained by the closed operation of morphology, and then the abnormal elevation point in the tree crown height model is replaced by the smooth elevation value through the normalization, binarization and convolution of the tree crown height model matrix. The low point between the crown and the crown remains, allowing successive crowns to be repaired and aligned. The optimized tree crown height model is separated by multi-scale single tree. Object-oriented multi-scale segmentation is a bottom-up region merging technique, which starts with a pixel object. Small image objects can be merged into larger objects. In this study, the tree crown height model is used to generate the image object prototype by using the object-oriented multi-scale segmentation method, and the region growth algorithm is realized by setting different scale parameters. The crown height model of the studied area is segmented and the ideal segmentation effect is obtained. Finally, by establishing the linear regression relationship between the characteristic values of trees estimated by radar and the measured values of trees, the tree height and crown width of individual trees are inversed. The average accuracy of the estimation of crown amplitude is 88%, the average precision of tree height estimation is up to 89%.
【学位授予单位】：中南林业科技大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TN958.98

【参考文献】