Multi-class Oil Palm Trees Condition Detection from UAV Images using Faster R-CNN with EfficientNetV2

In regions like Indonesia, known for extensive Oil Palm tree production, tree counting and tree condition detection accuracy are pivotal in assessing and forecasting the country's oil palm tree production. Faster R-CNN emerges as a deep learning method suitable for tree condition detection to increase detection precision. This paper proposes modifications to the backbone of Faster R-CNN to enhance tree condition detection performance, particularly when applied to UAV images for assessing tree conditions. In this study, the EfficientNetV2, especially the EfficientNetV2-S backbone, has stood out in performance for evaluation results and has become the utmost model in this study. The F1 Score from this model reached 80% and 73% for IoU@50 and IoU@75, respectively. This result is preferable to the other models used in this study.