基于改进YOLOv11n普洱龙珠茶外观质量检测分级算法

打开文本图片集

中图分类号：S571.1；TP391.4；TP18 文献标识码：A DOI：10.7525/j.issn.1006-8023.2026.01.019

Appearance Quality Detection Grading Algorithm for Pu'er Dragon Ball Tea Based on Improved YOLOviin

Jianchao，LI Wei*，TI Hailong，JIANG Chenxi，LIAO Hongsen，LI Jianlong （ ）

Abstract：Theapperance quality Pu'er Dragon Ballteaplays a decisiverole in itsmarket value；however，conventionalinspectionapproaches failtosimultaneouslysatisfythedemsreal-time eficiency，accuracy，edge-level deployment.In response，we propose SHM-YOLO，a lightweight object detection framework．Extending YOLOv11，the model employs ShufleNetV2（denoted as S in SHM）as the backbone，integrating point wise group convolution with chanel shufling to minimizecomputationalcost.Through the integrationahierarchicalscale featurepyramid network （HS-FPN，denoted as Hin SHM）thatcombineschannel atention withdimensional matching，the model strengthens the effctiveness multi-scale feature fusion.At thesame time，the multi-scale attention block（MAB，denotedas M in SHM）is utilized tooptimizethe C3K2 structure，enabling more effctive imagedetail extraction.To improve boundingbox regresion，the model combines Inner-IoU with SIoU loss，which expedites convergenceaugments localization precision.Experimental validationona self-developed dataset for Pu'er Dragon Balltea appearance qualityconfirms thatSHM-YOLO reaches 97.2% mAP@50， 92.7% precision （ P ）， 93.6% recall （R⋅ ），303 fps，withmerely0.969 1× 106 parameters 2.3 MB storage consumption.Compared to YOLOv11n，the model achieves higher accuracy while markedly decreasing floating-point computation （by 62.5% ） memory consumption（by 47.6% ），highlighting its excellentlightweight characteristics strong suitability for industrial deployment.

Keywords： Pu'er Dragon Balltea;appearance quality; improved YOLOv11n; ShufleNetv2; HS-FPN; MAB; high precision detection；lightweight model

0 引言

随着科技的不断进步和市场需求的日益增长，现代信息技术和智能农机装备在智慧农业领域的应用越来越广泛[1-2]。（剩余21590字）