<![CDATA[This study introduces various YOLOv8 architectures for detecting tree objects as a step in determining vegetation density using UAVs. The objective of this study is to evaluate the most optimal YOLO architecture that can be used to detect tree objects in dense forest vegetation in real-time by reviewing the model performance and the inference time required by the related model. Training was conducted using 2273 images with the following details: 2073 images for training, 67 images for testing, and 133 images for validation. For the best model, further training was carried out using a fine-tuning method to minimize reading errors made by the model. Based on training and testing, the most optimal model used for real-time forest monitoring is the YOLOv8n model with an F1-score value reaching 0.961 in testing at a height of 70 m. The most optimal flight altitude of the YOLO architecture for monitoring is at a height of 70 m, indicated by the highest F1-score value in the YOLOv8l architecture of 0.973. The use of the fine-tuning method has been proven to significantly reduce object reading errors by the YOLO architecture.]]>
Copyrights © 2026
