摘要
针对试管标签识别中的多字段、密排列,以及环境背景噪声和反光等干扰问题,提出一种基于改进YOLOv5s模型的视觉检测方法。利用OpenCV及优化后的CLAHE函数,对成像质量进行预处理;加入基于字段感知与多尺度强化的FA-SimAM预注意力层;将Res2Conv-RFEM混合模块替换C3部分模块,提升对字符层面的细粒度感知与密集信息处理能力,同时增强字段间的内在关联性;引入优化后的CBAM注意力机制,增强字段特征关注;用FA-NWD损失函数替换原CIoU损失函数,提升极小目标检测性能,并增强字段动态适配能力。结果表明,改进模型的mAP值为87.3%,召回率为87.5%,精确率为89.2%,相比原始YOLOv5s模型,分别提升12.3,7.2,7.1个百分点。研究为医疗识别相关领域提供支持。
To address the challenges in test tube label recognition,such as multiple densely-arranged fields,environmental background noise,and specular reflections,a visual detection method based on an improved YOLOv5s model is proposed.OpenCV and an optimized CLAHE function were utilized to preprocess image quality.A pre-attention layer based on field perception and multi-scale enhancement(FA-SimAM)was incorporated.The Res2Conv-RFEM hybrid module was used to replace parts of the C3 modules,thereby enhancing fine-grained perception at the character level and the ability to process dense information,while also strengthening the intrinsic relationships between fields.An optimized CBAM attention mechanism was introduced to increase focus on field features.The original CIoU loss function was replaced with the FA-NWD loss function,improving the detection performance for extremely small targets and enhancing dynamic adaptation capability for fields.The results show that the improved model achieved an mAP of 87.3%,a recall of 87.5%,and a precision of 89.2%,representing improvements of 12.3,7.2,and 7.1 percentage points,respectively,compared to the original YOLOv5s model.This research provides support for related fields in medical identification.
作者
金诚
芦金石
季旭
冯怡然
JIN Cheng;LU Jinshi;JI Xu;FENG Yiran(School of Mechanical Engineering and Automation,Dalian Polytechnic University,Dalian 116034,China;Key Laboratory for Seafood Processing Technology and Equipment of Liaoning Province,Dalian Polytechnic University,Dalian 116034,China)
出处
《包装与食品机械》
北大核心
2025年第6期19-29,共11页
Packaging and Food Machinery
基金
国家重点研发计划项目(2018YFD0400800)
辽宁省教育厅科研项目(JYTMS20230395)。
