Oil Palm Tree Detection using YOLOv8 Image Segmentation for Autonomous Robot

Accurate tree segmentation in unstructured outdoor environments is essential for autonomous navigation in agricultural robotics. To address this need, this study proposes a deep learning-based method for segmenting palm oil trees in unstructured outdoor environments. This study evaluates the performance of the YOLOv8 segmentation network in comparison with Mask R-CNN and DeepLabV3 for oil palm trees under real-world plantation conditions. Under natural field conditions, images were collected and annotated in instance segmentation format to train and evaluate deep learning models for oil palm tree detection. Performance was assessed based on accuracy, segmentation quality, and inference speed. Among the models tested, YOLOv8m demonstrated the best trade-off between segmentation accuracy and computational efficiency. It achieved the fastest inference time (0.0119 s per image), the shortest training time (3.36 hours), and maintained a competitive mean Pixel Accuracy (mPA) of 97.45%, a respectable mean Intersection over Union (mIoU) of 70.98%, and high precision at 81.90%. These results highlight YOLOv8m as the most suitable model for real-time segmentation in robotic applications operating in unstructured environments with limited computational resources.