计算机科学与探索 ›› 2022, Vol. 16 ›› Issue (1): 41-58.DOI: 10.3778/j.issn.1673-9418.2110003

• 综述·探索 • 上一篇    下一篇

深度学习中的单阶段小目标检测方法综述

李科岑1, 王晓强1,+(), 林浩2, 李雷孝3, 杨艳艳3, 孟闯3, 高静4   

  1. 1.内蒙古工业大学 信息工程学院,呼和浩特 010080
    2.天津理工大学 计算机科学与工程学院,天津 300384
    3.内蒙古工业大学 数据科学与应用学院,呼和浩特 010080
    4.内蒙古农业大学 计算机与信息工程学院,呼和浩特 010011
  • 收稿日期:2021-09-17 修回日期:2021-11-08 出版日期:2022-01-01 发布日期:2021-11-20
  • 通讯作者: + E-mail: wangxiaoqiang@imut.edu.cn
  • 作者简介:李科岑(1997—),女,山西人,硕士研究生,主要研究方向为深度学习、目标检测。
    王晓强(1978—),男,内蒙古人,副教授,硕士生导师,主要研究方向为大数据分析、智能交通。
    林浩(1995—),男,天津人,博士研究生,主要研究方向为网络安全、数据挖掘。
    李雷孝(1978—),男,山东人,博士,教授,主要研究方向为云计算、大数据分析、数据挖掘、目标检测。
    杨艳艳(1997—),女,山东人,硕士研究生,主要研究方向为云计算、大数据分析、计算机视觉。
    孟闯(1997—),男,山东人,硕士研究生,主要研究方向为人工智能、目标检测、图像处理、云计算、大数据分析。
    高静(1970—),女,博士,教授,主要研究方向为云计算、大数据分析、农业信息化。
  • 基金资助:
    内蒙古自治区关键技术攻关计划项目(2019GG273);内蒙古自治区科技成果转化专项资金项目(2020CG0073);内蒙古自治区科技成果转化专项资金项目(2021CG0033);内蒙古自治区科技重大专项(2019ZD015);内蒙古自治区科技重大专项(2019ZD016);内蒙古自治区科技计划项目(2020GG0104)

Survey of One-Stage Small Object Detection Methods in Deep Learning

LI Kecen1, WANG Xiaoqiang1,+(), LIN Hao2, LI Leixiao3, YANG Yanyan3, MENG Chuang3, GAO Jing4   

  1. 1. College of Information Engineering, Inner Mongolia University of Technology, Hohhot 010080, China
    2. College of Computer Science and Engineering, Tianjin University of Technology, Tianjin 300384, China
    3. College of Data Science and Application, Inner Mongolia University of Technology, Hohhot 010080, China
    4. College of Computer and Information Engineering, Inner Mongolia Agricultural University, Hohhot 010011, China
  • Received:2021-09-17 Revised:2021-11-08 Online:2022-01-01 Published:2021-11-20
  • About author:LI Kecen, born in 1997, M.S. candidate. Her research interests include deep learning and object detection.
    WANG Xiaoqiang, born in 1978, associate professor, M.S. supervisor. His research interests include big data analysis and intelligent transportation.
    LIN Hao, born in 1995, Ph.D. candidate. His research interests include network security and data mining.
    LI Leixiao, born in 1978, Ph.D., professor. His research interests include cloud computing, big data analysis, data mining and object detection.
    YANG Yanyan, born in 1997, M.S. candidate. Her research interests include cloud computing, big data analysis and computer vision.
    MENG Chuang, born in 1997, M.S. candidate. His research interests include artificial intelligence, object detection, image processing, cloud computing and big data analysis.
    GAO Jing, born in 1970, Ph.D., professor. Her research interests include cloud computing, big data analysis and agricultural informatization.
  • Supported by:
    Key Technology Research Program of Inner Mongolia(2019GG273);Project for Transformation of Scientific and Technological Achievements of Inner Mongolia(2020CG0073);Project for Transformation of Scientific and Technological Achievements of Inner Mongolia(2021CG0033);Major Science and Technology Project of Inner Mongolia(2019ZD015);Major Science and Technology Project of Inner Mongolia(2019ZD016);Science and Technology Plan Project of Inner Mongolia(2020GG0104)

摘要:

随着深度学习的不断发展,目标检测技术逐步从基于传统的手工检测方法向基于深度神经网络的检测方法转变。在众多基于深度学习的目标检测方法中,基于深度学习的单阶段目标检测方法因其网络结构较简单、运行速度较快以及具有更高的检测效率而被广泛运用。但现有的基于深度学习的单阶段目标检测方法由于小目标物体包含的特征信息较少、分辨率较低、背景信息较复杂、细节信息不明显以及定位精度要求较高等原因,导致在检测过程中对小目标物体的检测效果不理想,使得模型检测精度降低。针对目前基于深度学习的单阶段目标检测方法存在的问题,研究了大量基于深度学习的单阶段小目标检测技术。首先从单阶段目标检测方法的Anchor Box、网络结构、交并比函数以及损失函数等几个方面,系统地总结了针对小目标检测的优化方法;其次列举了常用的小目标检测数据集及其应用领域,并给出在各小目标检测数据集上的检测结果图;最后探讨了基于深度学习的单阶段小目标检测方法的未来研究方向。

关键词: 深度学习, 单阶段目标检测, 小目标检测

Abstract:

With the development of deep learning, object detection technology has gradually changed from traditional manual detection methods to deep neural network detection methods. Among many object detection algorithms based on deep learning, the one-stage object detection method based on deep learning is widely used because of its simple network structure, fast running speed and higher detection efficiency. However, the existing one-stage object detection methods based on deep learning do not have ideal detection results for small target objects in the detection process due to the lack of feature information, low resolution, complicated background information, unobvious details and higher positioning accuracy, which reduces the detection accuracy of the model. Aiming at the existing problems of one-stage object detection method based on deep learning, a large amount of one-stage small object detection technologies based on deep learning are studied. Firstly, the optimization methods for small object detection are systematically summarized from the aspects of Anchor Box, network structure, IoU (intersection over union) and loss function in the one-stage object detection methods. Secondly, the commonly used small object detection datasets and their application fields are listed, and the detection graphs on each small object detection dataset are given. Finally, the future research direction of one-stage small object detection methods based on deep learning is investigated.

Key words: deep learning, one-stage object detection, small object detection

中图分类号: