计算机工程与应用 ›› 2020, Vol. 56 ›› Issue (1): 25-37.DOI: 10.3778/j.issn.1002-8331.1910-0164

• 热点与综述 • 上一篇    下一篇

全卷积神经网络研究综述

章琳,袁非牛,张文睿,曾夏玲   

  1. 1.江西科技师范大学 数学与计算机科学学院,南昌 330038
    2.上海师范大学 信息与机电工程学院,上海 201418
  • 出版日期:2020-01-01 发布日期:2020-01-02

Review of Fully Convolutional Neural Network

ZHANG Lin, YUAN Feiniu, ZHANG Wenrui, ZENG Xialing   

  1. 1.School of Mathematics and Computer Science, Jiangxi Science and Technology Normal University, Nanchang 330038, China
    2.College of Information, Mechanical and Electrical Engineering, Shanghai Normal University, Shanghai 201418, China
  • Online:2020-01-01 Published:2020-01-02

摘要: 近年来,全卷积神经网络发展迅猛,在多个视觉研究领域表现出了非常亮眼的成绩。重点收集了近几年的高质量文献,对其中提出的全卷积方法进行分析总结,力求让读者通过对研读,对全卷积神经网络的关键技术、研究现状和最新进展有一个比较全面的了解。将收集到的文献,按照研究领域的不同进行分类汇总,重点提取几个研究非常活跃的领域,详细介绍一些非常具有代表性的算法,并重点介绍了各种方法的精髓所在,同时还对近一年来的最新研究进展进行了概述。通过对大量文献的梳理研究,总结出全卷积神经网络在近几年取得的成就,分析各种方法的优缺点,根据全卷积神经网络目前还存在的一些问题,归纳出未来可能的发展方向。

关键词: 全卷积神经网络, 卷积计算, 深度学习, 视觉研究

Abstract: In recent years, the fully convolutional neural network has been developing rapidly, and has shown very bright results in many visual research fields. This paper focuses on collecting a large amount of recent high-quality literatures, analyzing and summarizing the fully convolutional methods proposed, and trying to make the reader has a more comprehensive understanding of key technologies, research status and latest developments of the fully convolutional neural network through the study of the paper. The collected literatures are classified and summarized according to the different fields of study, it focuses on extracting several areas where research is very active, and introduces in detail some of the most representative and state-of-the-art classic algorithms, and highlights the quintessence of the various methods. It also provides an overview of the latest research progress in the past year. Through the research on a large number of literatures, the achievements of the fully convolutional neural network are summarized, and the advantages and defects of these methods are analyzed. Based on the existing problems in the fully convolutional neural network, the possible future development direction is given.

Key words: fully convolutional neural network, convolutional computation, deep learning, visual research