Work

P-YOLO11: An improved lightweight model for accurate detection of declining trees in poplar plantations

Year: 2025

Type: article

Abstract: Accurate identification of declining trees in poplar plantations is essential for the implementation of effective forest management strategies and the promotion of sustainable forestry development in the Three Norths region. However, the complex environment within poplar plantations poses substantial challenges for declining tree detection, such as the high similarity between trees and their background, variations in light intensity, and changing weather conditions. To enhance the recognition efficiency and accuracy of declining trees in poplar plantations, this study proposes an improved lightweight and efficient model, P-YOLO11, based on the YOLO11n architecture. The P-YOLO11 model replaces the original detection head of YOLO11 with a dynamic detection head incorporating deformable convolution operators, which significantly enhances the detection accuracy of declining trees. Furthermore, the integration of the CSWin Transformer module strengthens the model's feature extraction capability, thereby improving both detection accuracy and robustness under complex environmental conditions. Experimental results demonstrate that the P-YOLO11 model achieves superior performance in detecting declining trees in poplar plantations, attaining a detection accuracy of 82.1%, which markedly surpasses that of current mainstream lightweight models (including YOLOv5n, YOLOv6n, YOLOv8n, YOLOv9n, YOLOv10n, and YOLO11n). Additionally, while maintaining the lightweight characteristics of the model, P-YOLO11 achieves a 7.2% improvement in detection accuracy and a 1.5% increase in mAP compared to the original YOLO11 model, highlighting its remarkable real-time detection capability and suitability for resource-constrained environments. Overall, this study provides an efficient and practical technical solution for the detection of declining trees in poplar plantations and offers significant reference value for related research and applications.

Source: Smart Agricultural Technology

DOI: 10.1016/j.atech.2025.101454