Individual Tree Crown Delineation From UAS Imagery Based on Region Growing by Over-Segments With a Competitive Mechanism

Unmanned aerial systems (UAS) have become a flexible and low-cost platform to supplement aerial or satellite remote sensing for precision forestry. The data derived from UAS are widely used to measure variables at a single tree level where individual tree crowns become fundamental. Most research has adapted some region growing method for individual tree crown delineation (ITCD). However, pixels are often used as growing units without considering the spatial and contextual information, which can be adversely affected by noise (e.g., background or branches) in the imagery. Instead, over-segments can compensate for these pixels’ shortcomings while also partially detecting the edge of a tree crown. These over-segments then become the growing units used in this study. In addition, this research incorporated competition among the over-segments to alleviate the deficits of sequential ordering. The algorithm was evaluated in three study sites with distinctive forest patterns utilizing natural color imagery. Results demonstrated that using over-segments as growing units improved the ITCD accuracy by 1.8%–2.3%, whereas incorporating the competitive mechanism further increased the accuracy by 4.3%–9.3%. The spatial arrangement of trees also affected the segmentation accuracy. The sources of uncertainties, such as the manually interpreted treetops and feature selection for region growing, were also analyzed. The algorithm developed in this research can be easily extended to other data sources to achieve promising accuracy.