电子学报 ›› 2021, Vol. 49 ›› Issue (7): 1417-1427.DOI: 10.12263/DZXB.20200651

• 综述评论 • 上一篇    下一篇

基于深度学习的显著性目标检测方法综述

罗会兰1, 袁璞1, 童康2   

  1. 1.江西理工大学信息工程学院,江西 赣州 341000
    2.南京航空航天大学电子信息工程学院, 江苏 南京 211106
  • 收稿日期:2020-07-03 修回日期:2020-09-23 出版日期:2021-07-25 发布日期:2021-08-11
  • 作者简介:罗会兰 女,1974年9月生于江西上高.2008年获浙江大学工学博士学位.现为江西理工大学图像处理实验室教授、硕士生导师.主要从事机器学习、模式识别等方面的研究.E-mail:luohuilan@sina.com
    袁 璞 女,1997年5月生于江西吉安.2018年进入江西理工大学.在读硕士研究生,研究方向为图像修复、显著性目标检测. E-mail:emmanuel_97@163.com
    童 康 男,1992年生于江苏南京,博士研究生,主要研究方向为计算机视觉与模式识别.E-mail:tkangcv@nuaa.edu.cn
  • 基金资助:
    国家自然科学基金(61862031);江西省赣州市“科技创新人才计划”

Review of the Methods for Salient Object Detection Based on Deep Learning

Hui-lan LUO1, Pu YUAN1, Kang TONG2   

  1. 1.College of Information Engineering,Jiangxi University of Science and Technology,Ganzhou,Jiangxi 341000,China
    2.College of Electronic and Information Engineering,Nanjing University of Aeronautics and Astronautics,Nanjing,Jiangsu 100084,China
  • Received:2020-07-03 Revised:2020-09-23 Online:2021-07-25 Published:2021-08-11

摘要:

显著性目标检测旨在对图像中最显著的对象进行检测和分割,是计算机视觉任务中重要的预处理步骤之一,且在信息检索、公共安全等领域均有广泛的应用.本文对近期基于深度学习的显著性目标检测模型进行了系统综述,从检测粒度的角度出发,综述了将深度学习引入显著性目标检测领域之后的研究成果.首先,从三个方面对显著性目标检测方法进行了论述:稀疏检测方法,密集检测方法以及弱监督学习下的显著性目标检测方法.然后,简要介绍了用于显著性目标检测研究的主流数据集和常用性能评价指标,并对各类主流模型在三个使用最广泛的数据集上进行了性能比较分析.最后,本文分析了显著性目标检测领域目前存在的问题,并对今后可能的研究趋势进行了展望.

关键词: 显著性目标检测, 深度学习, 卷积神经网络, 视觉显著性, 弱监督学习, 计算机视觉任务

Abstract:

Salient object detection aims to detect and segment the most salient objects in the image. It is one of the important preprocessing steps in computer vision tasks, and it has a wide range of applications in information retrieval, public safety and other fields. This paper systematically reviews the recent research on the salient object detection models based on deep learning. From the perspective of detection granularity, the research results of applying deep learning into the field of salient object detection are reviewed. First, the salient object detection methods are discussed from three aspects: sparse detection methods, dense detection methods and weakly-supervised learning methods. Then, the mainstream data sets and common performance evaluation indicators used for salient object detection research are briefly introduced, and the performance of various mainstream models on the three most widely used data sets are compared and analyzed. Finally, this paper analyzes the current problems in the field of salient object detection and prospects for possible future research trends.

Key words: salient object detection, deep learning, convolutional neural network, visual saliency, weakly supervised learning, computer vision task

中图分类号: