High-Precision Agricultural Analysis via YOLOv9 and U-Net on Drone Imagery

This study presents a drone-based image processing approach for evaluating plant health and deformations in large agricultural fields. High-resolution images collected using a drone system were analyzed with the YOLOv9 object detection and UNet segmentation algorithms. While YOLOv9 provided rapid and effective object detection over wide areas, the U-Net segmentation model enabled detailed pixel-level analyses, facilitating the creation of precise health maps related to plant health. These maps, integrated with geographic information systems (GIS), served as a critical tool for accurately identifying regions requiring intervention. The results demonstrated that YOLOv9 reliably detected deformations with high performance metrics, including $79 \%$ accuracy, $78 \%$ precision, and $82 \%$ sensitivity. U-Net segmentation further refined these detections by precisely distinguishing healthy and problematic areas with high accuracy. The developed system not only enhances resource efficiency and reduces costs in agricultural operations but also makes significant contributions to environmental sustainability.