包装与食品机械2025，Vol.43Issue(3)：88-95,8.DOI:10.3969/j.issn.1005-1295.2025.03.010
基于改进YOLOv8n的再造烟叶原料缺陷检测方法研究
Research on defect detection method for reconstituted tobacco leaf raw materials based on improved YOLOv8n
摘要
Abstract
To address insufficient multi-scale representation capability and low detection efficiency in surface defect inspection of raw materials during slurry-processed reconstituted tobacco leaf production,an intelligent detection network based on an improved YOLOv8n architecture is proposed.The CSP-SDCV module was designed to replace the original C2f module,optimizing feature extraction efficiency.The ADown module was introduced to enhance multi-scale feature representation,while a lightweight shared convolutional detection head reduced parameter redundancy.Local window attention mechanism was incorporated to strengthen boundary sensitivity for occluded targets.Experimental results show the improved model achieved 98.1%mAP@50 on the tobacco leaf defect dataset,representing a 1.8 percentage point increase over baseline YOLOv8n,with parameter count and computational load reduced by 54.4%and 50.6%respectively.This research provides high-precision,low-resource solutions for automated quality inspection in tobacco industry.关键词
烟叶缺陷检测/多尺度特征融合/轻量化检测头/局部窗口注意力/YOLOv8nKey words
tobacco leaf defect detection/multi-scale feature fusion/lightweight detection head/local window attention/YOLOv8n分类
轻工纺织引用本文复制引用
刘雄斌,刘志昌,胡念武,姚建武,陈一桢,唐天明,王晚霞,陈寒..基于改进YOLOv8n的再造烟叶原料缺陷检测方法研究[J].包装与食品机械,2025,43(3):88-95,8.基金项目
湖北省科技创新人才计划项目(2023DJCO68) （2023DJCO68）
湖北省中央引导地方科技发展专项(2024EIA041) （2024EIA041）
