DFL-YOLO: Density-Guided Frequency Decoupling and Large Selective Kernels improve dense multi-class tree crown detection
- 1ollege of Information Engineering, Sichuan Agricultural University, Ya'an, China
Abstract
Multi-class tree crown detection in unmanned aerial vehicle (UAV) imagery is essential for forest monitoring but remains challenged by severe target-scale variation, feature ambiguity caused by crown overlap and background interference, and unstable bounding box regression in crowded scenes. To address these coupled errors, we propose Density-Guided YOLO with Frequency Decoupling and Large Selective Kernels (DFL-YOLO), a YOLOv11-based framework that combines scale-context construction, density-conditioned feature disambiguation, and improved bounding box regression within a progressive detection process. First, structurally pruned Large Selective Kernel (LSK) modules combine cascaded depthwise convolutions with spatial selection to enrich local and long-range context along selected backbone and neck paths. Second, our Density-Guided Frequency Decoupling Branch (DFDB) predicts a Gaussian-supervised density mask from the resulting context-enhanced P4 feature and applies complementary modulation to its high- and low-frequency components, preserving boundary-sensitive responses while suppressing background interference. The modulated P4 feature is then fused with the P3 and P5 paths and passed to the multiscale detection head. Finally, Multi-Strategy Weighted Intersection over Union (MW-IoU) supervises bounding box regression by combining MPDIoU geometric constraints, the non-monotonic focusing mechanism of Wise-IoU, and task-oriented soft weights. On the Mixed NEON dataset constructed from NEON and Roboflow imagery, DFL-YOLO achieved 82.2% mAP@50 and 54.7% mAP@50:95, representing absolute improvements of 2.6% and 4.2% over the YOLOv11l baseline, respectively. After dataset-specific retraining on VisDrone2019, DFL-YOLO achieved 54.5% mAP@50 and 33.7% mAP@50:95, showing that the architecture can be trained for another dense aerial-imaging domain. In general, DFL-YOLO is a reliable method for detecting multi-class tree crowns in intricate forest environments.
Related
- A global three-dimensional leaf functional space reveals partial decoupling between carbon investment and water regulation
- Soil intrinsic penetration (SIP) index: decoupling environmental noise from mechanical resistance to improve soybean yield prediction
- 4D sensitivity kernels for infrasound atmospheric propagation: considering the need and impact of time dependent observation operators for acoustic wave observations in data assimilation methods
- Decoupling Kinetic and Hydrological Controls on Global Silicate Weathering via a Stratified Observational Factorial Framework
- Decoupling air and water: how throughflow shapes mountain lake temperatures
BibTeX
@article{liu2026dflyolo,
title = {{DFL-YOLO: Density-Guided Frequency Decoupling and Large Selective Kernels improve dense multi-class tree crown detection}},
author = {Yuling Liu and Jiong Mu and Binhong Zhou and Junhao Chen and Xinyue Li and Chunyu Chen},
journal = {Frontiers in Forests and Global Change},
year = {2026},
month = oct,
doi = {10.3389/ffgc.2026.1964673},
url = {https://doi.org/10.3389/ffgc.2026.1964673},
note = {CC BY},
}
RIS
TY - JOUR
TI - DFL-YOLO: Density-Guided Frequency Decoupling and Large Selective Kernels improve dense multi-class tree crown detection
AU - Yuling Liu
AU - Jiong Mu
AU - Binhong Zhou
AU - Junhao Chen
AU - Xinyue Li
AU - Chunyu Chen
PY - 2026
DA - 2026/10/26
JO - Frontiers in Forests and Global Change
DO - 10.3389/ffgc.2026.1964673
UR - https://doi.org/10.3389/ffgc.2026.1964673
AB - Multi-class tree crown detection in unmanned aerial vehicle (UAV) imagery is essential for forest monitoring but remains challenged by severe target-scale variation, feature ambiguity caused by crown overlap and background interference, and unstable bounding box regression in crowded scenes. To address these coupled errors, we propose Density-Guided YOLO with Frequency Decoupling and Large Selective Kernels (DFL-YOLO), a YOLOv11-based framework that combines scale-context construction, density-conditioned feature disambiguation, and improved bounding box regression within a progressive detection process. First, structurally pruned Large Selective Kernel (LSK) modules combine cascaded depthwise convolutions with spatial selection to enrich local and long-range context along selected backbone and neck paths. Second, our Density-Guided Frequency Decoupling Branch (DFDB) predicts a Gaussian-supervised density mask from the resulting context-enhanced P4 feature and applies complementary modulation to its high- and low-frequency components, preserving boundary-sensitive responses while suppressing background interference. The modulated P4 feature is then fused with the P3 and P5 paths and passed to the multiscale detection head. Finally, Multi-Strategy Weighted Intersection over Union (MW-IoU) supervises bounding box regression by combining MPDIoU geometric constraints, the non-monotonic focusing mechanism of Wise-IoU, and task-oriented soft weights. On the Mixed NEON dataset constructed from NEON and Roboflow imagery, DFL-YOLO achieved 82.2% mAP@50 and 54.7% mAP@50:95, representing absolute improvements of 2.6% and 4.2% over the YOLOv11l baseline, respectively. After dataset-specific retraining on VisDrone2019, DFL-YOLO achieved 54.5% mAP@50 and 33.7% mAP@50:95, showing that the architecture can be trained for another dense aerial-imaging domain. In general, DFL-YOLO is a reliable method for detecting multi-class tree crowns in intricate forest environments.
SN - 2624-893X
ER -