Multi-Class Oil Palm Tree Detection from UAV Imagery Using Deep Learning
Southeast Asia (SEA) region leads the world in palm oil production, with Indonesia, Malaysia, and Thailand collectively contributing over 88% of the global production. However, the tropical climate in the SEA region resulted in oil palm trees vulnerable to various diseases such as Fatal Yellowing (FY) and Ganoderma boninense. To keep track of productivity, it is crucial to monitor the varying conditions of oil palm trees—such as healthy, dead, yellow, and small—and apply effective pruning techniques to cure affected trees. Manual approach for oil palm tree detection is expensive, tedious, and prone to inaccuracies. Thus, our study is focused to automate the detection of oil palm trees and their states using a deep learning (DL) algorithm on unmanned aerial vehicle (UAV) imagery. We use YOLOv8, one of the latest open-source models, on the publicly available UAV dataset, named MOPAD, containing training and validation sets. The performance of the model is further compared with other state-of-the-art object detectors. In addition, a prototype web application is developed to demonstrate the robustness of the model in adverse real-world conditions and for potential deployment.