ABSTRACT
Airborne-based LiDAR has many advantages over the passive remote sensing for the extraction of a single tree canopy and the development of an accurate inventory of forestry. The traditional watershed algorithm exposes challenges due to the over-segmentation and sensitivity to noise. This chapter presents the K-means clustering watershed algorithm for single tree segmentation. This method includes: (1) eliminating the gross values (outliers), such as ground elevation, which heavily impact LiDAR point cloud data segmentation; (2) generating a canopy height model (CHM) from the LiDAR point cloud; (3) generating the cluster management module (CMM) from the CHM using the variable window detection method; (4) obtaining the treetop position for the following K-means cluster segmentation; (5) using the K-means clustering algorithm for initial cluster segmentation to extract the target pixels of interest using the local maximum value detected by the variable window; and (6) applying the watershed algorithm with four neighborhoods to segment the target image, after which the over-segmentation regions are merged, and the contour of the single tree canopy is identified. The experimental results demonstrated that the proposed algorithm can effectively overcome the over-segmentation of the traditional watershed algorithm.
