计算机工程与应用 ›› 2018, Vol. 54 ›› Issue (17): 1-9.DOI: 10.3778/j.issn.1002-8331.1806-0377

• 热点与综述 • 上一篇    下一篇

深度卷积神经网络在目标检测中的研究进展

姚群力1,2,胡  显1,2,雷  宏1   

  1. 1.中国科学院 电子学研究所,北京 100190
    2.中国科学院大学,北京 100049
  • 出版日期:2018-09-01 发布日期:2018-08-30

Application of deep convolutional neural network in object detection

YAO Qunli1,2, HU Xian1,2, LEI Hong1   

  1. 1. Institute of Electronics, Chinese Academy of Sciences, Beijing 100190, China
    2. University of Chinese Academy of Sciences, Beijing 100049, China
  • Online:2018-09-01 Published:2018-08-30

摘要: 深度卷积神经网络以多层次的特征学习与丰富的特征表达能力,在目标检测领域取得了突破进展。概括了卷积神经网络在目标检测领域的研究进展,首先回顾传统目标检测的发展及存在的问题,引出卷积神经网络的目标检测基本原理和基本训练方法;然后分析了以R-CNN为代表的基于区域建议的目标检测框架,介绍以YOLO算法为代表的将目标检测归结为回归问题的目标检测框架;最后,对目前目标检测的一些问题进行简要总结,对未来深度卷积神经网络在目标检测的发展进行了展望。

关键词: 深度卷积神经网络, 目标检测, 特征表达, 特征提取

Abstract: Deep convolutional Neural Networks(DNNs) have made breakthroughs in object detection, because of its more powerful ability of feature learning and feature representation. In this paper, the research progress of convolutional neural networks has been expounded in object detection. Firstly, the development and existing problems of traditional object detection are reviewed. It introduces the principles of the DNNs and the improvement of common techniques. Then, object detection framework which combines region proposal is introduced. It introduces YOLO algorithm, which aims at object detection as regression problem. Finally, some existing problems in the present study are briefly summarized and the new directions for future development are expected.

Key words: Deep convolutional Neural Networks(DNNs), object detection, feature representations, feature extraction