高精度实时语义分割算法框架: 多通道深度加权聚合网络
CSTR:
作者:
作者单位:

1. 内蒙古工业大学 电力学院,呼和浩特 010080;2. 大规模储能技术教育部工程研究中心, 呼和浩特 010080;3. 内蒙古自治区高等学校智慧能源技术与装备工程研究中心, 呼和浩特 010080;4. 北京工业大学 信息学部,北京 100080

作者简介:

通讯作者:

E-mail: qys@imut.edu.cn.

中图分类号:

TP183

基金项目:

国家自然科学基金项目(62241309);内蒙古科技计划项目(2020GG028,2021GG164);内蒙古自然科学基金项目(2020MS05029,2021MS06018).


High precision real-time semantic segmentation algorithm: Multi-channel deep weighted aggregation network
Author:
Affiliation:

1. School of Electric Power, Inner Mongolia University of Technology, Hohhot 010080, China;2. Engineering Research Center of Large Energy Storage Technology of Ministry of Education, Hohhot 010080, China;3. Center for Intelligent Energy Technology and Equipment Engineering, Inner Mongolia University, Hohhot 010080;4. Faculty of Information Technology, Beijing University of Technology, Beijing 100080, China

Fund Project:

  • 摘要
  • |
  • 图/表
  • |
  • 访问统计
  • |
  • 参考文献
  • |
  • 相似文献
  • |
  • 引证文献
  • |
  • 资源附件
  • |
  • 文章评论
    摘要:

    近年来随着深度学习技术的不断发展,涌现出各种基于深度学习的语义分割算法,然而绝大部分分割算法都无法实现推理速度和语义分割精度的兼得.针对此问题,提出一种多通道深度加权聚合网络(MCDWA_Net)的实时语义分割框架.\:该方法首先引入多通道思想,构建一种3通道语义表征模型,3通道结构分别用于提取图像的3类互补语义信息:低级语义通道输出图像中物体的边缘、颜色、结构等局部特征;辅助语义通道提取介于低级语义和高级语义的过渡信息,并实现对高级语义通道的多层反馈;高级语义通道获取图像中上下文逻辑关系及类别语义信息.\:之后,设计一种3类语义特征加权聚合模块,用于输出更完整的全局语义描述.\:最后,引入一种增强训练机制,实现训练阶段的特征增强,进而改善训练速度.\:实验结果表明,所提出方法在复杂场景中进行语义分割不仅有较快的推理速度,且有很高的分割精度,能够实现语义分割速度与精度的均衡.

    Abstract:

    In recent years, with the continuous development of deep learning technology, various semantic segmentation algorithms based on deep learning have emerged, but most of the segmentation algorithms cannot achieve high speed and high accuracy at the same time, and a real-time semantic segmentation framework for multi-channel depth-weighted aggregation networks (MCDWA_Net) is proposed to solve this problem. Firstly, the multi-channel idea is introduced to construct a three-channel semantic representation model, which is used to extract three types of complementary semantic information of the image: 1) Low-level semantic channel outputs the local features such as the edge, color, and structure of the object in the image; 2) Auxiliary semantic channel extracts the transition information between low-level semantics and high-level semantics, and realizes multi-layer feedback to the high-level semantic channel; 3) Advanced semantic channel obtains context logical relationships and category semantic information in images. Then, a three-class semantic feature weighted aggregation module is designed to output a more complete global semantic description. Finally, an enhancement training mechanism is introduced to realize the feature enhancement in the training stage, thereby improving the training speed. Experimental results show that the proposed method not only has fast inference speed, but also has high segmentation accuracy in complex scenes, which can achieve the balance of semantic segmentation speed and accuracy.

    参考文献
    相似文献
    引证文献
引用本文

齐咏生,陈培亮,高学金,等.高精度实时语义分割算法框架: 多通道深度加权聚合网络[J].控制与决策,2024,39(5):1450-1460

复制
分享
文章指标
  • 点击次数:
  • 下载次数:
  • HTML阅读次数:
  • 引用次数:
历史
  • 收稿日期:
  • 最后修改日期:
  • 录用日期:
  • 在线发布日期: 2024-04-17
  • 出版日期: 2024-05-20
文章二维码