| Forest ecosystems which is produced by the interactions between forests and the environment is the cornerstone of human life in water conservation,sand fixation and play an important role in the social and economic development.In the early 1980s,China has already implemented comprehensive monitoring of forest resources and established a data collection,analysis and evaluation management system.The validity,timeliness and accuracy of the investigation of forest resources were becoming inevitable.However,traditional methods of manual investigation of forest resources monitoring have the limitation of low precision,low efficiency and small coverage,which fail to fulfill the requirements of modem forest monitoring.With the rapid development of hardware equipment and the improvements in remote sensing technology,UAV and other fields,precise and efficient monitoring of forest resources has been realized.The combination of airborne LiDAR system and high resolution CCD camera provides a new method of data collection.It can obtain high resolution remote sensing images and the 3D coordinates of the forest at the same time,which can further be applied to extract information such as the tree height,crown and tree density.Therefore,extraction of forest parameters based on high-resolution images and LiDAR point cloud has caught increasing attention.However,the improvement in sampling density of point cloud is accompanied by the increased input cost,and the limitations of remote sensing technology in tree height extraction.In order to extract parameters in a wide range of individual tree with high efficiency and low cost,and avoid the error accumulation during the experiment,this paper proposed a method for the extraction of high and individual tree parameters of individual tree canopy.The author selected the DaYeKou forest in ZhangYe city,Gansu Province as a test area and adopted low sampling density point cloud data acquisition and synchronization of CCD high resolution image data.For high-resolution CCD images,the author proposed a segmentation method for individual tree crown based on the gray gradient image.Comparing Roberts or Laplace operator and the Improved mathematical morphology operators,the author found the optimal operator is improved mathematical morphology operators.Then,the author used mathematical morphological operators and improved object-oriented multi-scale segnentation method to rapidly extract the information of a large range of individual tree crown.The results show that the classification accuracy of trees quantity is 83.19%,and the accuracy of trees shape reaches 88.62%using the extraction of individual tree crown from the gray gradient and high resolution image by object oriented segmentation.Both the two precisions are higher than the precision of traditional forestry survey.Secondly,in order to estimate trees height,the author classified the raw LIDAR data by distance cluster analysis method based on shape indices.And then the classification of the point cloud is executed by mathematical filter and inverse interpolation morphology filter.By comparing with the measured data,the precise of trees height reaches 87.88%.Points cloud was interpolated to generate DEM and DSM.The author obtained the CHM by grid computing of DEM and DSM.The individual tree height 1s estimated by the combination of CHM and canopy vector.The precise of trees9 height 1s 82.32%,meaning the individual tree height extraction can be realized by a low sampling density.In summary,based on the gray level gradient image segmentation method,the author can obtain the complete information of crown edge,and rapidly extract individual tree crown in a wide range.Although the forest vertical structure is complex and the point cloud data has a low sampling density,the author demonstrate the feasibility and necessity of individual tree height extraction by distance clustering method based on shape index. |
PDF Full Text Request