Research on tree species identification method based on YOLOv8 and visible light images

In the field of forest resource management and biodiversity surveys, traditional manual survey methods are gradually being replaced by efficient automated technologies. This study utilizes UAV (unmanned aerial vehicle) visible light imagery to investigate the effectiveness of the YOLOv8 model in classifying tree species in natural forests. The research focuses on the natural secondary forest in Mao'ershan Forest Farm, constructing a dataset of seven tree species. Subsequently, a tree species instance segmentation model was established based on the YOLOv8 algorithm to segment tree crowns and predict tree species. By optimizing strategies and analyzing convergence, Radam was identified as the optimal optimizer, with the model converging after 200 training epochs and achieving an mAP50 of 89.6%. In summary, this study proposes a novel method for tree species recognition based on UAV visible light imagery combined with the YOLOv8 network. Using the tree species identification in the natural secondary forest of Mao'ershan Forest Farm as an example, the validity of the instance segmentation model was demonstrated.