基于航片密集匹配影响因素分析的树高提取

发布时间：2018-04-18 11:00

本文选题：历史航片 + 激光雷达　；参考：《中国林业科学研究院》2017年硕士论文

【摘要】：本研究对密集匹配的各种影响因素进行分析,集成多时期历史航片与机载激光雷达数据监测树高生长。同时,首先基于激光雷达点云数据构建林下高精度数字高程模型与数字表面模型,然后利用不同航向角与不同分辨率的航片,构建不同的航片组合方案,再构建立体像对,通过自动立体匹配算法生成密集匹配数字表面模型。借助数字高程模型将数字表面模型进行高度归一化,提取研究区域的森林冠层高度。以此验证空间分辨率、航向角、影像覆盖率、地形、阴影对密集匹配的影响。再利用1996年、2004年的历史航片与2014年的数字航片和激光雷达数据,构建了皖南杉木林18年内三个时期的森林冠层高度。论文主要结果如下:(1)经基于面片的三维多视角(PMVS)、改进半全局(tSGM)、半全局匹配(SGM)等密集匹配算法产生的森林冠层高度,与激光雷达观测得到的冠层高度拟合后得到的相关性R2分别是0.71、0.76、0.64,RMSE分别是2.17m、3.5m、2.16m,SGM具有较高的精度;(2)由不同航向角组合的航片经密集匹配得到的森林冠层高度与激光雷达数据得到的冠层高度的相关性R2分别为0.49、0.65、0.52、0.76、0.56,RMSE分别为2.86m、3.19m、1.96m、2.16m、3.81m,xxx组合具有较高的精度;(3)由2014年数字航片处理得到的森林冠层高度与对应样地实测数据验证精度较高,最大绝对误差为3.53m,平均绝对误差为1.5m,最大相对误差为31.29%,平均相对误差为14.27%,测量精度为85.73%;(4)由1996年、2004年、2014年航片得到的三期森林冠层高度增长趋势与杉木树高生长曲线趋势一致。数字航片经不同的密集匹配算法对于测量冠层高度的能力是相似的,但是在冠层细节表现能力与高程精度上各有侧重。在多山复杂地形条件下,提高航片空间分辨率对于冠层细节的获取具有提升作用,但是不能提升森林冠层高度的精度,其精度的提升主要依靠增加不同航向角的航片。利用历史航片能准确定量反映山脊向阳面的森林冠层高度变化,对于山谷阴影处,则会出现冠层高度被低估情况。利用多期航片结合高精度DEM数据可定量反映研究区优势木冠层高度变化,为多时期历史航片监测森林生长趋势和评价林地生产力提供可能。
[Abstract]:In this study, the various factors affecting the dense matching were analyzed, and the tree height growth was monitored by integrating multi-period historical aerial photographs and airborne lidar data.At the same time, the high precision digital elevation model and digital surface model under forest are constructed based on the point cloud data of lidar, and then different combination schemes of aerial pictures are constructed by using different heading angles and different resolutions, and then stereo image pairs are constructed.The dense matching digital surface model is generated by the automatic stereo matching algorithm.The height of the digital surface model is normalized by the digital elevation model and the forest canopy height of the study area is extracted.The effects of spatial resolution, heading angle, image coverage, terrain and shadow on dense matching are verified.The canopy height of Chinese fir forest in South Anhui Province during the past 18 years was constructed by using the historical aerial photographs of 1996 and 2004 and the digital aerial and lidar data of 2014.The main results of this paper are as follows: (1) the forest canopy height is generated by the improved semi-global and semi-global matching algorithms, such as multi-angle PMVSs based on surface slices, etc.The correlation R ~ 2 obtained by fitting the canopy height obtained by Lidar is 0.71 ~ 0.766 ~ 0.64m RMSE is 2.17mt ~ (3.5m) ~ (2.16mSGM) respectively, and the forest canopy height is obtained by dense matching between the canopy height and the Lidar by the combination of different heading angles.The correlation R2 of the canopy height obtained from the data is 0.49 / 0. 65 / 0. 52N / 0. 726 / 0. 56 RMSE = 2. 86mg / r = 3.19 mU / 1.96 m / L / 2.16mGN / 3.81mxxx respectively. The forest canopy height obtained from the digital aerial photo processing in 2014 has a higher accuracy than that obtained from the data measured in the corresponding sample plots, and the accuracy of verification is higher than that of the data obtained from the digital aerial photograph processing in 2014, and the correlation between the canopy height and the data measured in the corresponding samples is higher.The maximum absolute error is 3.53m, the average absolute error is 1.5m, the maximum relative error is 31.29m, the average relative error is 14.27m, and the measuring accuracy is 85.73cm) the growth trend of forest canopy height obtained in 1996, 2004 and 2014 is consistent with that of Chinese fir tree height growth curve.The different dense matching algorithms of digital aerial photographs are similar to the ability to measure the height of the canopy, but there are different emphases on the performance of the canopy details and the elevation accuracy.Under the condition of multi-mountain and complex terrain, improving the spatial resolution of aerial film can improve the canopy details, but it can not improve the accuracy of forest canopy height. The improvement of the accuracy mainly depends on the increase of aerial photographs with different heading angles.The historical aerial photographs can accurately and quantitatively reflect the variation of forest canopy height from ridge to the sun, and for the shadow of valley, the canopy height will be underestimated.Multi-period aerial photographs combined with high-precision DEM data can be used to quantitatively reflect the canopy height changes of dominant trees in the study area, which provides the possibility for monitoring forest growth trends and evaluating forest productivity by multi-period historical aerial photographs.
【学位授予单位】：中国林业科学研究院
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TN958.98;S758

【参考文献】