Research On Forest Single Tree Identification Based On Airborne LiDAR

As a new earth observation technology,Light Detection And Ranging(LiDAR)has been widely applied in many fields,such as forest resource inventory,digital city,power line inspection and basic surveying and mapping.Compared with traditional optical remote sensing,LiDAR can overcome the limation of signal saturation and cloud effect,penetrate through the forest canopy to the ground,and thereafter directly obtain the structural information of forest canopy.Thus it has a unique advantage in describing the spatial structure of forest canopy.Among all LiDAR techiniques,airborne LiDAR can directly and efficiently obtain the three-dimensional structure information of forest canopy with high precision and high-density because of its smaller light spot diameter and the full digital recording echo information(echo times and echo intensity).In this case,it is suitable for accurate detection of single tree position and precise segmentation of tree crown.Overall,airborne LiDAR can provide data source support for quantitative and fine description of forest three-dimensional structure,such as tree height,crown,leaf area index,etc.Additioanlly,it has therefore become one of the most effective techiniques of forest resource inventory.In view of the low detection precision of the existing single-tree segmentation algorithm based on airborne LiDAR data in dense forest area,the single-tree segmentation algorithm for dense forest areas was studied in this research.Several aspects content are included in this paper as following:(1)the K-dimension tree was used to improve the search efficiency of point cloud data.Then,the noise was removed,and mathematical morphology filtering algorithm,moving surface filtering algorithm and cloth simulation filtering algorithm were used to classify the ground point and the non ground point from the laser point cloud data.(2)Ground points and the unground points were interpolated respectively to generate the digital elevation model(DEM)and the digital surface model(DSM)by the inverse distance weighting method;based on DSM and DEM,the canopy height model(CHM)was generated and optimized.At the same time,the canopy point cloud normalization was conducted.(3)An effective method was proposed to better detect individual tree from airborne LiDAR data.The proposed method can directly detects individual trees from point cloud by fully considering the characteristics of vertical stratification in the canopy space.First,the initial segmentation of the canopy was achieved by the normalized cut(Ncut)segmentation with a prior knowledge of single-tree positions derived from the local maximum of the CHM.Finally,the Ncut method was used again to reduce the leakage rate of individual trees.Different from the second step,this step used the global maximum rather than the local maximum as a prior knowledge of single-tree position.Additionally,the shape and the minimum point number of the canopy were also taken into consideration as constraints in this step.The experimental results show that the proposed method effectively improves the accuracy of tree delineation,up to 90%in high-density forest.(4)In order to verify the reliability and advantages of the proposed method,the segmentation results before and after unchecked single-tree processing were evaluated quantitatively,and compared with the results of other methods.Experiments show that this method has advantages and potentials,and will contribute to quantitative,detailed description and parameter inversion of three dimensional structure in single-tree scale.