基于改进YOLOv7-tiny的轻量级条码检测算法  

Lightweight Barcode Detection Algorithm Based on Improved YOLOv7-tiny

作　　者：王正家[1,2] 丁聪 庄健 肖喆 程培 杨剑东 WANG Zheng-jia;DING Cong;ZHUANG Jian;XIAO Zhe;CHENG Pei;YANG Jian-dong(School of Mechanical Engineering,Hubei University of Technology,Wuhan 430068,China;Key Laboratory of Modern Manufacture Quality Engineering,Hubei University of Technology,Wuhan 430068,China)

机构地区：[1]湖北工业大学机械工程学院,武汉430068 [2]湖北工业大学现代制造质量工程湖北省重点实验室,武汉430068

出　　处：《印刷与数字媒体技术研究》2025年第1期71-81,共11页Printing and Digital Media Technology Study

基　　金：国家自然科学基金项目(No.51275158)。

摘　　要：针对当前复杂工业场景下条码检测精度低、多尺度识别难度大、检测算法复杂度高的问题,本研究提出一种基于改进YOLOv7-tiny的轻量级条码检测算法。首先,针对检测算法复杂度高、难部署到嵌入式设备的问题,引入ShuffleNet v2轻量化网络并将其结构中步长为2的深度可分离卷积修改为空洞卷积来扩大感受野,修改后作为新的特征提取网络。其次,嵌入CBAM(Convolutional Block Attention Module)轻量级注意力机制提高网络特征提取能力,获取更丰富的语义信息,提升小目标检测精度。最后,采用SIoU损失函数替代原始的CIoU损失函数,增强条码定位能力。实验结果表明,改进后的YOLOv7-tiny模型相比原模型的平均精度和速度分别提升了2.36%和19frame/s、参数量和计算量分别减少了0.9MB和1.9G,满足工业场景下条码检测准确度与速度的要求。Aiming at the current problems of low barcode detection accuracy,high difficulty in multi-scale recognition,and high complexity of detection algorithms in complex industrial scenarios,a lightweight barcode detection algorithm based on improved YOLOv7-tiny was proposed in this study.Firstly,in order to solve the problem that the detection algorithm is highly complex and difficult to deploy to embedded devices,the ShuffleNet v2 lightweight network was introduced.The depth-separable convolution with a stride of 2 in its structure was modified into an atrous convolution to expand the receptive field,which became a new feature extraction network.Secondly,the lightweight attention mechanism of CBAM(Convolutional Block Attention Module)was embedded to improve the network feature extraction capability,obtain richer semantic information,and improve small target detection accuracy.Finally,the SIoU loss function was used to replace the original CIoU loss function,which enhanced barcode positioning capabilities.The experiment results showed that compared with the original model,the average accuracy and speed of the improved YOLOv7-tiny model are increased by 2.36%and 19frame/s respectively,and the number of parameters and calculations are reduced by 0.9MB and 1.9G respectively.The improved YOLOv7-tiny model meets the requirements of barcode detection accuracy and speed in industrial scenarios.

关 键 词：条码 深度学习 目标检测 轻量级 YOLOv7-tiny 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]