| It is a hot study field to extract forest structure parameter using Airborne LiDAR.The forest stem height plays a significant role in all the parameters and is the basis for inverting estimates of other forest parameters.At present,a lot of research has focused on the extraction of forest stem height information from the LiDAR data.However,most of those only focus on high-density LiDAR data due to the tree height estimated in this way is influenced profoundly by the sample density and the forest fragmentation,and lack of some study on the low-density LiDAR data.Although,the joint use of different remotely sensed sources could provide a good solution for improving the performances of the low-density LiDAR data,only few sdudies investigated the problem of estimating forest stem height considering the joint use of low-density LiDAR data instead of using high-density LiDAR and optical images,especially the high resolution digital camera images.And also,these studies do not report accurate estimates since none of them focused the attention on the precise estimation of the tree top height at single tree level.In order to address this issue,in this paper we propose a data fusion approach to the estimation of forest stem height at single tree level and plot-level of conifer forest that effectively exploits low-density LiDAR data and digital camera images having high geometric resolution.Main content and conclusion is follows:1)The filtering algorithm of the LiDAR data on the complex forest district was discussed.The progressive TIN filter algorithm was analyzed especially.The experimental results demonstrate that this filtering algorithm can better adapt to the forest area compared with others,and keep topographical features better.The mean square error(MSE)of the DEM was maintained at about 0.15 meter to meet the needs of forestry investigation.2)Based on the 24 laser-derived features,which derived from the vegetation point dataset,and the filed data,the estimation model for the random forest regression of the mean canopy height in the study areas was established.The result showed that the average value of the estimated tree height has a significant linear correlation with the measured values,and the correlation coefficient was above 96%.The accuracies of all plots were higher than 87%and the total average accuracy was 93.17%.3)Individual tree crown information was extracted from high-resolution optical images.In this paper,we proposed an improved marker-controlled watershed segmentation algorithm to accurately extract a single tree crown polygon.Second,an improved local maximum method was proposed to detecting the positions of individual trees in digital camera images,which is about 0.5 meter away from the field measured canopy position.Final,Individual tree crown diameter was estimated by using an improved valley-following method,and the estimation accuracy was above 75%,the average estimate error was controlled in 0.3 meters.4)In this paper,we propose a 3-dimensional parametric model of the tree to the estimation of the tree top height at single tree level based on the joint use of Individual tree crown derived from the optical image and the height information provided by the LiDAR data.Compared with the tree height directly from the low-density LiDAR data,the accuracies was better than 80%and the root mean square error(RSME)of estimation was about 0.8 meters.The experimental results show that the proposed method can effectively solve the problem that the low-density LiDAR data can't obtain the information of the single tree level,and greatly expand the use value and range of the LiDAR data. |
PDF Full Text Request