CD-YOLO: A lightweight end-to-end detection model for cigarette appearance defects

Appearance defect detection is essential for ensuring cigarette quality during production. Reaching high-precision and lightweight automated cigarette appearance defect detection has long been manufacturers' key focus. However, existing methods struggle to balance detection accuracy and speed effectively. This paper proposes a high-performance detection model for cigarette defects, named cigarette defect YOLO (CD-YOLO), which builds upon the YOLOv10 network with three major improvements. First, an intra-scale feature interaction (ISFI) module is designed to enhance the model's ability to distinguish different defects. Subsequently, a multi-scale feature fusion (MSFF) network is developed to improve the model's performance in recognizing small-scale and subtle defects. Finally, a lightweight group convolution detection head (LGCDH) is implemented to substantially reduce the model's computational complexity and parameter count, accelerating detection speed. The experimental results demonstrate that the CD-YOLO model achieves a favorable trade-off between accuracy and speed, maintaining a detection speed exceeding 500 FPS, with a mAP@0.5 of 96.2%. Additionally, a novel data augmentation strategy is introduced in this paper, employing low-rank adaptation (LoRA) to fine-tune a pretrained stable diffusion model, which generates synthetic defect samples to alleviate data scarcity.