YOLOv8-CTB:Lightweight and Real-Time Cotton Bud Detection under Field Conditions with Edge Deployment

In the field cotton planting scenario, challenges such as high planting density, complex backgrounds, and limited hardware resources make the accurate identification of cotton apical buds extremely difficult. To tackle this problem, this paper constructs a lightweight detection model named YOLOv8-CTB..The proposed model incorporates three key optimisations: (1) The ExtraDW module was introduced to reconstruct the C2f structure, enhancing feature extraction capability while reducing computational redundancy; (2) a lightweight cross-scale module (BF-CCFM), inspired by BiFPN, to enhance multi-scale feature fusion, thereby improving detection robustness; and (3) the SEAM attention mechanism, which strengthens the model’s ability to focus on critical features in occluded regions, mitigating detection errors. Experimental results show that YOLOv8-CTB achieves a detection accuracy of 98.3% on the self-built cotton apical bud dataset, which is 2.3 percentage points higher than YOLOv8n. Meanwhile, FLOPs, parameter count, and model size are reduced to 74.1%, 75.9%, and 47.1% of the benchmark model, respectively.Furthermore, this study investigates the feasibility of deploying the YOLOv8-CTB model on mobile edge devices. After TensorRT acceleration, the model was deployed on a Jetson Xavier NX development board, where it achieved a recognition accuracy of 91.2% for cotton terminal buds. The findings suggest that the proposed model is highly applicable to cotton terminal bud detection under dense planting conditions, thereby providing a theoretical foundation for advancing intelligent topping practices.