Enhancing sustainability in the production of palm oil: creative monitoring methods using YOLOv7 and YOLOv8 for effective plantation management
Istiyak Mudassir Shaikh, Mohammad Nishat Akhtar, Abdul Aabid, Omar Shabbir Ahmed
Abstract
The You Only Look Once (YOLO) deep learning model iterations-YOLOv7-YOLOv8-were put through a rigorous evaluation process to see how well they could recognize oil palm plants. Precision, recall, F1-score, and detection time metrics are analyzed for a variety of configurations, including YOLOv7x, YOLOv7-W6, YOLOv7-D6, YOLOv8s, YOLOv8n, YOLOv8m, YOLOv8l, and YOLOv8x. YOLO label v1.2.1 was used to label a dataset of 80,486 images for training, and 482 drone-captured images, including 5,233 images of oil palms, were used for testing the models. The YOLOv8 series showed notable advancements; with 99.31 %, YOLOv8m obtained the greatest F1-score, signifying the highest detection accuracy. Furthermore, YOLOv8s showed a notable decrease in detection times, improving its suitability for comprehensive environmental surveys and in-the-moment monitoring. Precise identification of oil palm trees is beneficial for improved resource management and less environmental effect; this supports the use of these models in conjunction with drone and satellite imaging technologies for agricultural economic sustainability and optimal crop management.