摘要
概述机器视觉注意力机制的研究背景、发展中的显著工作及原理,综述软注意力机制中空间注意力机制、通道注意力机制、混合注意力机制和硬注意力机制,讨论Transformer的模型,并在分析各类改进模型的思路及特点的同时重点梳理了近3年的研究动态,最后探讨了注意力机制的在超参数优化、胶囊网络、神经架构搜索、多模态方面的潜在研究方向的应用展望.
This paper first outlines the research background,notable work in development and principles of machine vision attention mechanism,then reviews the spatial attention mechanism,channel attention mechanism,hybrid attention mechanism and hard attention mechanism in soft attention mechanism,also discusses Transformer′s model,and analyzes the ideas and characteristics of various types of improved models while focusing on combing the research dynamics in the last three years,and finally at the end of the paper discusses the potential research directions of attention mechanism in data dependence,capsule networks,neural architecture search,multimodality.
作者
冯小丹
云利军
高海峰
孟凤菊
FENG Xiao-dan;YUN Li-jun;GAO Hai-feng;MENG Feng-ju(School of Information Technology,Yunnan Normal University,Kunming 650500,China;Engineering Research Center of Computer Vision and Intelligent Control Technology,Department of Education of Yunnan Province,Kunming 650500,China)
出处
《云南民族大学学报(自然科学版)》
2025年第4期453-463,共11页
Journal of Yunnan Minzu University(Natural Sciences Edition)
基金
云南师范大学2023年度研究生科研创新基金(YJSJJ23-A17)。
