3D remote sensing of plant community using scanning lidar systems

利用高空间分辨率的直升机载扫描激光雷达系统对森林面积进行了三维测量。测试了激光雷达系统在几种受惊条件下的精度，并对系统的飞行条件进行了优化。结果表明，在均方根误差小于20 cm的情况下，可以较准确地估计出木本树高。接下来，研究了利用激光雷达数据自动检测树木高度和位置的算法。将新设计的算法(冠状提取滤波：CE滤波)与传统的分水岭算法和局部极大值滤波(LM滤波)进行了比较。结果表明，数字冠层高度模型(DCHM)能更准确地估计树高，平滑后用CE滤波识别树顶。结果表明，该方法识别树顶的误差为4.9%，树高的识别误差为-0.25~0.42m，均方根误差为0.38m，并设计了一种算法来识别每棵树的轮廓，从而绘制出森林生物的空间分布图；另一方面，利用便携式扫描激光雷达系统测量了森林中的树木，并估算了每棵树的胸径。根据胸径数据，估算了森林生物量。这种方法能够在不干扰林下和砍伐树木的情况下估计森林生物量。此外，从地面多个点测量的便携式激光雷达数据被合并到一张3D图像中，以补充盲区。然后，利用计算机图形学技术生成了几棵树的完整3D模型。从三维模型出发，计算了水平断面上的树高、树干直径、最大冠层直径和冠层面积以及冠层体积等参数，并对误差进行了评估。

Forest area was measured 3-dimensionally using helicopter-borne scanning lidar system with high spatial resolution. Accuracy of the lidar system at several fright conditions was examined and flight condition of the system could be optimized. As a result, woody tree height could be estimated accurately with RMSE less than 20 cm. Next, algorithms for automatic detection of tree height and position were examined using the lidar data. A newly designed algorithm (Crown Extraction filtering : CE filtering) was compared with the conventional ones, such as Watershed method and Local maximum filtering(LM filtering). It was shown that tree height was estimated more accurately from Digital Canopy Height Model(DCHM) with tree tops identified by CE filtering after smoothing. As a result, error of identification of tree tops using this method was 4.9 % and the one of tree heights ranged -0.25 to 0.42 m with RMSE of 0.38 m. Moreover, we designed an algorithm to identify outline of each tree and it enabled to map spatial distribution of forest biomass.On the other hand, understory of trees was measured in forest using a portable scanning lidar system and the diameter at breast height(DBH) of each tree was estimated. From the data of DBH, the forest biomass was estimated. This method enabled to estimate forest biomass without disturbing understory and felling trees. In addition, potable lidar data of trees measured from several points on the ground were merged into one 3D image to complement blind regions. Then, a complete 3D model of several trees was generated by a technique of computer graphics. From the 3D model, several parameters such as tree height, stem diameter, maximum canopy diameter and canopy area in the horizontal cross-section, and canopy volume were computed and the errors were evaluated.