Individual Tree Crown Segmentation in Subtropical Broadleaf Forests Using UAV-based Ultrahigh-Resolution RGB Data

Rapid and accurate extraction of individual tree crown information is the basis for fine investigation and scientific management of forest resources. However, in subtropical broad-leaved forests, the crown contains multiple tree apexes, and the commonly used watershed segmentation algorithms have different degrees of over-segmentation. In this study, we propose an individual-tree segmentation method with watershed segmentation combined with spectral-texture-intensity-controlled normalized cutting, which directly utilizes the DSM and DOM data generated from the high-resolution images of unmanned aerial vehicles (UAVs). The effectiveness of our method for individual-tree crown segmentation was verified in a subtropical broadleaf forest in southern China (F=0.87, R=0.93, P=0.82). The results show that our method can reduce the over-segmentation of watershed segmentation and provide an efficient and low-cost new method for crown information extraction in complex forest environments.