.png){kind=icon}
.png){kind=icon}
Return
RP-YOLO11: Lightweight and High-Accuracy Object Detector via Attention-Enhanced Architecture for the Detection of Red Pine Seedlings
.png){kind=icon}
OA
AI
R
Z
DOI：10.1109/access.2026.3718135.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Post-fire red pines (Pinus brutia) afforestation is crucial for the ecological sustainability and recovery of forest ecosystems. The regeneration of red pine seedlings can occur through natural processes or human-assisted planting, which often leads to imbalance in forest densities. As this situation negatively affects the ecological balance, it is essential to accurately determine and monitor seedling density. This paper proposes an improved red pine seedlings detection model based on YOLO11, called RP (Red Pine)-YOLO11. The proposed model is optimized to have a lighter model structure compared to the default YOLO11 architecture, and high-cost modules are removed from the architecture and more efficient modules such as C3k2+SA (Shuffle Attention) and C3k2_PSA (Position-Sensitive Attention) are integrated. As a result of these structural modifications, the model exhibits improved detection accuracy alongside enhanced computational efficiency. To verify the effectiveness of proposed model, we created a dataset consisting of 1587 images called RP-Dataset. Additionally, in order to address challenges associated with low-resolution imagery, super-resolution technique is employed to enhance detail learning. The proposed model demonstrates improvements of 1.3% and 3.3% in mAP50 and mAP50-95 metrics, respectively, compared to the baseline YOLO11. Moreover, the model achieves a significant reduction in complexity with approximately 4.6 times fewer parameters, 1.2 times lower GFLOPs and 3.4 times lower in model size. Additionally, performance has been further optimized through the implementation of transfer learning and fine-tuning strategies. Following transfer learning and fine-tuning, the model reaches a mAP50 of 97.6% and a mAP50-95 of 72.8%. The results indicate that RP-YOLO11 is a lightweight, efficient, and accurate model for small object detection in complex environments such as forests and extensive agricultural areas.
Keywords:
Computer vision
object detection
YOLO11
pine tree detection