Extraction of individual tree parameters of Chinese pine by oblique photogrammetry

  Objective  It is an important content in the process of forest management to obtain the information of single wood parameters. Oblique photogrammetry has become one of the most efficient methods to obtain single wood information with its multi-angle shooting method.

  Method  In this study, taking Chinese pine forest in Wangyedian, Inner Mongolia of northern China as the research object, the tree height, crown width and stem volume were obtained by UAV tilt photography, and the effects of four different photo resolutions (1, 0.5, 0.25, 0.1 m) on the information extraction ability of single tree were investigated. The mean-shift algorithm based on the point cloud data and the watershed algorithm based on CHM were used to segment the single tree crown. Taking the measured single tree parameters of the sample plot and the single tree data extracted by LiDAR as the verification data, the relationship between the resolution of photo and the extraction ability of single tree was explored, and the accuracy and the optimal resolution of the two segmentation methods were compared. The allometric model (y = 0.000 \; 1x^2.717, R² = 0.571 7) was established to extract tree height and volume with CHM, and the volume distribution map of Chinese pine in the survey area was drawn. According to the experimental results, the key parameters of oblique photogrammetry, such as the reasonable resolution requirements of UAV photos and the flight altitude range were obtained.

  Result  The watershed algorithm and mean-shift algorithm had the best segmentation accuracy at 0.5 m photo resolution, and the canopy extracted by watershed algorithm had less missing points and over segmentation than mean shift algorithm, with F-scores of 0.87 and 0.82, respectively. While at the resolution of 0.5 m, the crown value extracted by mean-shift algorithm was more accurate. The correlation coefficients between the reference crown and the segmented crown obtained by watershed algorithm and mean-shift algorithm were 0.850 and 0.892, respectively, which were very significant at the level of 0.01. The ability of watershed segmentation algorithm and mean-shift algorithm to extract the height of Chinese pine were similar. The average error of the height of single tree obtained from 0.5 m photo resolution was the smallest and the difference was not big, which were 0.42 and 0.66 m, respectively.

  Conclusion  In this study, the key method and the optimum photo resolution parameters for extracting individual Chinese pine by oblique photogrammetry were defined, which improved the investigation efficiency and provided scientific basis for setting reasonable data acquisition parameters of UAV.