中国烟草学报2025，Vol.31Issue(5)：31-40,10.DOI:10.16472/j.chinatobacco.2025.T0043
基于改进YOLOv8 的烟叶碎片中杂物的识别
Identification of debris in tobacco leaf fragments based on an improved YOLOv8 model
摘要
Abstract
[Purpose]This study aimed to improve the removal efficiency of debris in the threshing and redrying process of tobacco leaves and to address challenges such as complex scene interference and missed detection of small targets.[Methods]A debris detection method based on an improved YOLOv8 model was proposed.A multi-branch coordinate attention fusion module(MBCAF)was designed to enhance multi-scale feature extraction,and a large-kernel convolutional attention mechanism(LKASPPF)was introduced to optimize global feature capture.In addition,an extra detection head was added to reduce the missed detection rate of small targets.[Results]After improvement,the model achieved a mean average precision(mAP)of 98.5%,representing a 15.1%increase compared with the baseline YOLOv8 model,while significantly reducing the missed detection rate of small targets.Furthermore,validation on real-world production lines confirmed that the model enables high-precision,real-time debris detection and removal under complex production environments.[Conclusion]The proposed method effectively realizes debris detection and removal in the centralized processing line of threshing and redrying fragments,providing a feasible solution for automatic impurity removal and intelligent production in tobacco leaf processing.关键词
烟叶分拣/杂物检测/YOLOv8/多分支坐标注意力Key words
tobacco sorting/debris detection/YOLOv8/multi-branch coordinate attention(MBCAF)引用本文复制引用
楚晗,童帅帅,张相辉,韩校星,徐帅华,郭坤,张攀峰,党霞..基于改进YOLOv8 的烟叶碎片中杂物的识别[J].中国烟草学报,2025,31(5):31-40,10.基金项目
天昌国际烟草有限公司重点项目"打叶复烤非烟物质智能剔除技术研究"(2023410003440013) （2023410003440013）
