计算机科学与探索 ›› 2022, Vol. 16 ›› Issue (5): 1025-1042.DOI: 10.3778/j.issn.1673-9418.2111063

• 综述·探索 • 上一篇    下一篇

深度卷积应用于目标检测算法综述

董文轩, 梁宏涛(), 刘国柱, 胡强, 于旭   

  1. 青岛科技大学 信息科学技术学院,山东 青岛 266061
  • 收稿日期:2021-11-11 修回日期:2022-01-24 出版日期:2022-05-01 发布日期:2022-05-19
  • 通讯作者: + E-mail: lht@qust.edu.cn
  • 作者简介:董文轩(1997—),男,山东德州人,硕士研究生,CCF学生会员,主要研究方向为机器学习、计算机视觉等。
    梁宏涛(1979—),男,山东济宁人,博士,副教授,CCF高级会员,主要研究方向为数据挖掘、能源互联网、计算机视觉等。
    刘国柱(1965—),男,山东青岛人,硕士,教授,主要研究方向为网络安全、大数据、计算机视觉等。
    胡强(1980—),男,山东邹城人,博士,副教授,CCF专业会员,主要研究方向为软件形式化验证、服务计算、计算机视觉等。
    于旭(1982—),男,山东青岛人,博士,副教授,CCF专业会员,主要研究方向为推荐系统、迁移学习、计算机视觉等。
  • 基金资助:
    国家自然科学基金(61973180);山东省产教融合研究生联合培养示范基地项目(2020-19)

Review of Deep Convolution Applied to Target Detection Algorithms

DONG Wenxuan, LIANG Hongtao(), LIU Guozhu, HU Qiang, YU Xu   

  1. School of Information Science and Technology, Qingdao University of Science and Technology, Qingdao, Shandong 266061, China
  • Received:2021-11-11 Revised:2022-01-24 Online:2022-05-01 Published:2022-05-19
  • About author:DONG Wenxuan, born in 1997, M.S. candidate, student member of CCF. His research interests include machine learning, computer vision, etc.
    LIANG Hongtao, born in 1979, Ph.D., associate professor, senior member of CCF. His research interests include data mining, Internet of energy, computer vision, etc.
    LIU Guozhu, born in 1965, M.S., professor. His research interests include network security, big data, computer vision, etc.
    HU Qiang, born in 1980, Ph.D., associate professor, professional member of CCF. His research interests include software formal verification, service computing, computer vision, etc.
    YU Xu, born in 1982, Ph.D., associate professor, professional member of CCF. His research interests include recomendation system, transfer learning, computer vision, etc.
  • Supported by:
    National Natural Science Foundation of China(61973180);Industry-Education Integration Postgraduate Joint Cultivation Demonstration Base Project of Shandong Province(2020-19)

摘要:

目标检测作为计算机视觉中最基本、最具挑战性的任务之一,旨在找出图像中特定的目标,并对目标进行定位和分类,现已被广泛应用于工业质检、视频监控、无人驾驶等众多领域。近年来,随着计算机硬件资源和深度卷积算法在图像分类任务中取得突破性进展,基于深度卷积的目标检测算法也逐渐替代了传统的目标检测算法,在精度和性能方面取得了显著成果。综述了基于深度卷积的目标检测算法的研究现状以及今后可能的发展方向。以传统目标检测算法存在的局限性为引,首先介绍了目标检测算法权威的数据集和评估指标;再以时间和算法架构为研究主线,综述了近年来基于深度卷积的目标检测代表性算法的研究和发展历程,对比分析了单阶段、双阶段以及其他改进算法的网络架构,并归纳总结出各类目标检测算法所存在的特点、优势和局限;最后结合当下目标检测存在的问题与挑战对未来趋势进行展望。

关键词: 计算机视觉, 深度卷积, 目标检测, 单阶段, 双阶段

Abstract:

As one of the most fundamental and challenging tasks in computer vision, target detection aims to find out specific targets in images and to locate and classify them, and is now widely used in many fields such as industrial quality inspection, video surveillance and unmanned vehicles. In recent years, with the breakthroughs in computer hardware resources and depth convolution algorithms in image classification tasks, depth convolution-based target detection algorithms have gradually replaced the traditional target detection algorithms and achieved significant results in terms of accuracy and performance. This paper reviews the current research status of depth convolution-based target detection algorithms and possible future development directions. It introduces the authoritative datasets and evaluation metrics of target detection algorithms with the limitations of traditional target detection algorithms as a guide, and then reviews the research and development history of representative algorithms for depth convolution-based target detection in recent years with time and algorithm architecture as the main research lines. The network structures of one-stage, two-stage and other improved algorithms are compared and analyzed, and the characteristics, advantages and limitations of various target detection algorithms are summarized. Finally, the future trends are prospected in the light of current problems and challenges of target detection.

Key words: computer vision, deep convolution, target detection, one-stage, two-stage

中图分类号: