Detecting Red Pine Seedlings with YOLOv8: A Labeling Method Comparison

The detection and periodic monitoring of small pine seedlings are crucial for effective forest management. However, the presence of weeds in the field makes it difficult to detect small pine seedlings, necessitating the use of new techniques to address the challenge. The success of the techniques used in the field of object detection depends on various factors. The labeling technique used for the dataset is among the most important of these factors. In this work, we examine the impact of different data labeling techniques on the success of the YOLOv8 object detection algorithm in detecting small pine seedlings. Within the scope of the work, we created a dataset by labeling drone images taken from the Mersin Yeşilovacık forest area using both polygon and bounding box methods and analyzed the performance of YOLOv8 Large and Nano models. According to the results, the polygon method performed better than the bounding box method. Additionally, the YOLOv8 Large model is superior to the Nano model in both labeling methods.