烟草科技2026，Vol.59Issue(5)：66-76,11.DOI:10.16135/j.issn1002-0861.2025.0538
基于改进YOLOv11n的密集烤房烟叶烘烤状态识别模型
A recognition model for tobacco leaf curing status in bulk curing barns based on improved YOLOv11n
摘要
Abstract
To address the issues of low recognition accuracy and weak generalization ability for tobacco leaf curing status due to gradual color change and vein overlapping,a recognition model TOC-YOLO for tobacco leaf curing status in bulk curing barns based on the improved YOLOv11n was proposed.The model enhanced the extraction of tobacco leaf morphological features by introducing the VanillaNet module.In addition,by embedding the SENetV2 module and the BiFormer attention mechanism to strengthen the visual perception of color features and capture of spatial features of tobacco leaves,the model's ability to understand complex spatial structures and occlusion relationships was improved.The model adopted the HS-FPN module to achieve multi-scale feature fusion,enhancing its adaptability to the dynamic morphological changes of tobacco leaves during curing,and integrated the ASFF module to reduce the misjudgment rate of adjacent curing status.The results showed that:1)In the test set,the bounding box precision(Bp),recall(R),F1 score and mAP0.50-0.95 of the TOC-YOLO model were 96.9%,96.1%,96.5%and 98.2%respectively,and its comprehensive performance was superior to comparative models such as YOLOv11n.2)Under the same experimental conditions,each of the improvement measure promoted the performance of the TOC-YOLO model,achieving favorable experimental results.3)In the recognition of various curing status,the average F1 score of the TOC-YOLO model was 96.5%,and the average mAP0.50-0.95 was 98.2%,indicating strong adaptability to different curing status.4)The TOC-YOLO model exhibited excellent feature perception ability in key curing stages such as the late yellowing stage,late color-fixing stage and late stem-drying stage.5)On the generalization test set,compared with the baseline YOLOv11n model,the Bp,R,F1 score and mAP0.50-0.95 of the TOC-YOLO model were increased by 5.3,2.5,4.0 and 0.8 percentage points respectively,demonstrating strong cross-scenario generalization capability.This study provides technical support for the recognition of tobacco leaf curing status in bulk curing barns.关键词
烟叶烘烤/状态识别/YOLOv11n/特征融合/注意力机制Key words
Tobacco leaf curing/Status recognition/YOLOv11n/Feature fusion/Attention mechanism分类
轻工纺织引用本文复制引用
张烨,陈铸荣,黎树杰,岑罗泽,霍煜东,李圣陶,孙懿清,李成杰..基于改进YOLOv11n的密集烤房烟叶烘烤状态识别模型[J].烟草科技,2026,59(5):66-76,11.基金项目
国家重点研发计划课题"主粮作物移动式烘干机专用传感器与智能装备创制"(2024YFD2000104) （2024YFD2000104）
国家自然科学基金项目"基于㶲分析法的稻谷深床干燥能量优化及其模型预测控制研究"(32401725) （32401725）
国家自然科学基金项目"粮食干燥准则及工艺过程解析理论研究"(32171906) （32171906）
广东省烟草专卖局科技项目"基于水分迁移和图谱特征的烟叶调制在线监测系统研究与应用"(2021440000240143). （2021440000240143）
