基于改进YOLOv8s的车轮毂背腔文字识别  

Character recognition of wheel hub cavity based on improved YOLOv8s

作　　者：朱梓清 陈小中[1] 杨淼[1] ZHU Ziqing;CHEN Xiaozhong;YANG Miao(Changzhou Institute of Engineering Technology,Changzhou 213001,China)

机构地区：[1]常州工程职业技术学院,常州213001

出　　处：《国外电子测量技术》2025年第1期94-102,共9页Foreign Electronic Measurement Technology

摘　　要：针对汽车轮毂背腔文字检测任务,小目标和复杂背景干扰是影响检测精度的关键因素,提出一种改进的YOLOv8s算法,通过设计高效卷积模块和引入注意力机制改善特征提取能力,从而提升检测精度。主要贡献包括:(1)引入可变自适应卷积(VAKConv)替代传统卷积操作,提高网络对多尺度目标的感知能力;(2)结合多头高效注意力(Efficient Multi-head Attention,EMA)机制重构C2f模块,形成C2f-BE模块,增强通道间的全局特征交互。通过消融实验和对比实验验证了算法性能,结果表明:改进算法在小目标检测和复杂背景处理上表现出色,平均精度(mAP)@50和mAP@50:95分别达到87.3%和52.4%,显著优于原始YOLOv8s。同时,该方法兼顾模型轻量化设计,具备较高的工程应用价值。Small targets and complex background interference are the key factors affecting the accuracy of the vehicle hub back cavity text detection task.In response to the above problems,this paper proposes an improved YOLOv8s algorithm,which improves the feature extraction ability by designing efficient convolution modules and introducing an attention mechanism to improve the detection accuracy.The main contributions of this paper include:(1)introducing variable adaptive convolution(VAKConv)to replace the traditional convolution operation to improve the perception ability of the network for multi-scale objects;(2)The C2f module is reconstructed by combining the efficient multihead attention(EMA)mechanism to form the C2F-BE module,which enhances the global feature interaction between channels.The results show that the improved algorithm performs well in small target detection and complex background processing.mAP@50 and mAP@50:95 reach 87.3%and 52.4%respectively,which are significantly better than the original YOLOv8s.At the same time,the proposed method considers the model’s lightweight design,which has high engineering application value.

关 键 词：可变自适应卷积 多头注意力 改进YOLOv8s 轮毂背腔字符 字符识别 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]