Morphology-guided dual-level detection and depth localization of cotton terminal buds for precision topping
- 1College of Mechanical and Electrical Engineering, Shihezi University, Shihezi
- 2Key Laboratory of Northwest Agricultural Equipment, Ministry of Agriculture and Rural Affairs, Shihezi
Abstract
Reliable terminal-bud localization is essential for precision cotton topping, yet the small bud core is often obscured by apical leaves and visually confounded with the surrounding canopy. We developed DGQ-YOLO, a lightweight morphology-guided detector that jointly identifies the bud core and a composite terminal-bud region encompassing the core and adjacent apical leaves. This dual-level representation preserves structural cues when the bud core is partly occluded. The model combines C2f_DWR_CA for multiscale, position-sensitive feature extraction, VoVGS-CA for lightweight feature fusion, and QCN loss for quality-aware classification and small-object regression. DGQ-YOLO was evaluated using 2,230 field images of long-staple cotton at the early bud-development stage. It achieved mAP@50 values of 93.2% and 94.1% for the bud core and composite terminal-bud region, respectively. Overall mAP@50 and mAP@50:95 were 93.6% and 81.7%, respectively. Compared with YOLOv8n, DGQ-YOLO improved overall mAP@50 by 2.2 percentage points. Its parameter count decreased from 3.01 to 2.90 million, and its computational cost fell from 8.2 to 7.8 GFLOPs. Combining the centers of the two detected regions with RGB-D depth measurements supported online terminal-bud localization in field experiments. This morphology-informed perception framework supports precision topping under the tested cultivar, developmental stage, and field conditions, although broader agronomic and actuator-level validation is required.
Related
- Projection-dependent VGluT1 and VGluT2 expression and terminal morphology in mouse visual circuits
- Design of a comprehensive protection algorithm for fault detection, classification, and location in high voltage direct current transmission lines connected to wind farms based on local terminal measurements
- Hybrid local–global vision transformer framework for robust plant disease detection in precision agriculture
- Chili-Seg: a high-precision detection and instance segmentation model for chili flowers in complex environments
- A real-time deep learning method for detection and localization of non-emerged cabbage seedlings
BibTeX
@article{wang2026morphologyguided,
title = {{Morphology-guided dual-level detection and depth localization of cotton terminal buds for precision topping}},
author = {Chaoqun Wang and Haojun Wen and Zhaohui Zheng and Haixiao Gou and Jie Zhang and Zhuoyan Zhang},
journal = {Frontiers in Plant Science},
year = {2026},
month = oct,
doi = {10.3389/fpls.2026.1966512},
url = {https://www.frontiersin.org/articles/10.3389/fpls.2026.1966512/full},
note = {CC BY 4.0},
}
RIS
TY - JOUR
TI - Morphology-guided dual-level detection and depth localization of cotton terminal buds for precision topping
AU - Chaoqun Wang
AU - Haojun Wen
AU - Zhaohui Zheng
AU - Haixiao Gou
AU - Jie Zhang
AU - Zhuoyan Zhang
PY - 2026
DA - 2026/10/14
JO - Frontiers in Plant Science
DO - 10.3389/fpls.2026.1966512
UR - https://www.frontiersin.org/articles/10.3389/fpls.2026.1966512/full
AB - Reliable terminal-bud localization is essential for precision cotton topping, yet the small bud core is often obscured by apical leaves and visually confounded with the surrounding canopy. We developed DGQ-YOLO, a lightweight morphology-guided detector that jointly identifies the bud core and a composite terminal-bud region encompassing the core and adjacent apical leaves. This dual-level representation preserves structural cues when the bud core is partly occluded. The model combines C2f_DWR_CA for multiscale, position-sensitive feature extraction, VoVGS-CA for lightweight feature fusion, and QCN loss for quality-aware classification and small-object regression. DGQ-YOLO was evaluated using 2,230 field images of long-staple cotton at the early bud-development stage. It achieved mAP@50 values of 93.2% and 94.1% for the bud core and composite terminal-bud region, respectively. Overall mAP@50 and mAP@50:95 were 93.6% and 81.7%, respectively. Compared with YOLOv8n, DGQ-YOLO improved overall mAP@50 by 2.2 percentage points. Its parameter count decreased from 3.01 to 2.90 million, and its computational cost fell from 8.2 to 7.8 GFLOPs. Combining the centers of the two detected regions with RGB-D depth measurements supported online terminal-bud localization in field experiments. This morphology-informed perception framework supports precision topping under the tested cultivar, developmental stage, and field conditions, although broader agronomic and actuator-level validation is required.
SN - 1664-462X
ER -