Abstract:

Objective: In individual tree segmentation using three-dimensional point could data, common problems include insufficient detection of the understory trees, the low proportion of true positive detections which decreases the effectiveness of the extracted information, and the detection being sensitive to the point density and the complexity of the forest structure. This study aimed to solve these problems by improving the segmentation algorithm, which was expected to support the implementation of the algorithm in the practice. Method: In this study, we proposed an algorithm for the detection and delineation of individual trees in heterogeneous and dense forests based on multi-slice clustering by using low density point cloud data, and the procedures for slicing, detection, segmentation and matching were also improved. Result: The results showed that the improved algorithm could achieve the detection and delineation of individual trees in heterogeneous and dense forests. The obtained trees could be reasonably matched with field trees, where the proportion of accurately extracted trees was 88.70%. Furthermore, the accuracy of individual tree height and mean height were 92.38% and 99.84%, respectively. The highest accuracy of the forest-structure parameters was 89.65%. Conclusion: In general, the main accomplishments of this study were as follows: 1) through the improvement of multi-slice clustering to the algorithm, we enhanced the detection capacity for understory and regeneration trees. 2) Through the establishment of an effectiveness index, named validity index, we were able to assess the effectiveness of the detection and delineation results. 3) Through the addition of the stand spatial structure parameters, we were able to utilize the prominent capacity of LiDAR in obtaining vertical structure information.

Key words: airborne LiDAR, individual tree segmentation, effectiveness of the information extraction, forest structure