Answers are grounded in this paper and cite their sources.
Seedling number is a pivotal metric for evaluating crop production potential. Conventional ground-based assessment is labor-intensive and prone to subjective inaccuracies. Although unmanned aerial vehicles enable scalable monitoring, accurately detecting small and overlapping faba bean seedlings under complex soil backgrounds remains challenging for existing object detection models. To address this issue, multiple input forms, including raw image, HSV and CIELab transformations, were systematically evaluated, and raw image achieved the best performance. Eight augmentation strategies expanded the original 961 UAV images to 8,649 annotated samples, enhancing robustness under variable field conditions. A comprehensive benchmark across the YOLOv5, YOLOv8 and YOLO11 families indicated that YOLO11s provided the optimal balance between accuracy and efficiency. Based on this backbone, FB-YOLO11s was developed by integrating a P2 detection head, a C3k2_MSCB2 module and an SENetV2 attention mechanism. The proposed model achieved mAP 50 = 0.9500 and F1-score = 0.9177 for faba bean seedling detection. The estimated and measured values presented strong agreement with an R 2 of 0.9444 and an RMSE of 8.4611. In addition, the FB-YOLO11s also achieved effective detection on pea seedlings (mAP 50 = 0.9059 and F1-score = 0.8824), indicating promising adaptability for broader early-stage crop monitoring, with the reported metrics evaluated on an augmented test set. This study could provide a practical and scalable solution for precision agriculture.
No takes yet. Share an insight, caveat, or question.
Jin et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: