基于多尺度特征的条形码快速检测算法  

Fast barcode detection algorithm based on multi-scale convolutional features

作　　者：易帆 李功燕[1] 许绍云[1] YI Fan;LI Gong-yan;XU Shao-yun(R&D Center for Intelligent Manufacturing Electronics,Institute of Microelectronics of Chinese Academy of Sciences,Beijing 100029,China;School of Electronic,Electrical and Communication,University of Chinese Academy of Sciences,Beijing 100049,China)

机构地区：[1]中国科学院微电子研究所智能制造电子研发中心,北京100029 [2]中国科学院大学电子电气与通信工程学院,北京100049

出　　处：《计算机工程与设计》2020年第1期220-225,共6页Computer Engineering and Design

基　　金：中国科学院弘光专项基金项目(KFJ-HGZX-012)

摘　　要：为提升在复杂环境下智能物流分拣系统中条形码检测的精度和速度,提出一种基于多尺度特征的条形码快速检测算法。采用深度学习中主流one-stage目标检测器作为基础框架,通过级联不同特征融合层和压缩层实现语义信息充分提取,在不同特征提取层分别嵌入膨胀卷积和深度可分离卷积,对特征提取效果和速度进行有效优化提升。将算法应用于实际分拣现场数据进行测试分析,与已有的YoLo-v3和Vgg-SSD网络等进行对比,该算法在准确度和速度方面具有明显优势,能够较好解决实际应用问题。To improve the accuracy and speed of barcode detection in intelligent logistics sorting system under complex environment,a fast barcode detection algorithm based on multi-scale features was proposed.The mainstream one-stage target detector in deep learning was adopted as the basic framework.The semantic information was fully extracted by cascading different feature fusion layers and compression layers,and the dilated convolution and depthwise separable convolution were embedded in different feature extraction layers,effectively optimizing the feature extraction effect and speed.The algorithm was applied to the actual sorting site for test analysis,and compared with the existing YoLo-v3 and Vgg-SSD networks.The algorithm has obvious advantages in accuracy and speed,which can better solve the actual application problem.

关 键 词：条形码检测 多尺度特征 卷积神经网络 特征融合 快速检测 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]