作者投稿和查稿 主编审稿 专家审稿 编委审稿 远程编辑

计算机工程 ›› 2021, Vol. 47 ›› Issue (10): 314-320. doi: 10.19678/j.issn.1000-3428.0058993

• 开发研究与工程应用 • 上一篇    

超尺度自导注意力网络的遥感船舶识别

陈会伟1, 刘树美1, 刘培学1, 公茂法2   

  1. 1. 青岛黄海学院 智能制造学院, 山东 青岛 266427;
    2. 山东科技大学 电气与自动化工程学院, 山东 青岛 266427
  • 收稿日期:2020-07-20 修回日期:2020-09-25 发布日期:2020-11-02
  • 作者简介:陈会伟(1982-),女,副教授、硕士,主研方向为控制工程、目标识别;刘树美,讲师、硕士;刘培学,副教授、硕士;公茂法,教授。
  • 基金资助:
    山东省重点研发计划(2017GGX20100,2019GGX105001);山东省高等学校科技计划项目(J18KB163);青岛黄海学院重点项目(2019KJ01,2019KJ02)。

Remote Sensing Ship Recognition Based on Hyper-Scale Self-Guided Attention Networks

CHEN Huiwei1, LIU Shumei1, LIU Peixue1, GONG Maofa2   

  1. 1. Institute of Intelligent Manufacturing, Qingdao Huanghai University, Qingdao, Shandong 266427, China;
    2. College of Electrical Engineering and Automation, Shandong University of Science and Technology, Qingdao, Shandong 266427, China
  • Received:2020-07-20 Revised:2020-09-25 Published:2020-11-02

摘要: 传统多尺度卷积神经网络因接收域有限,难以对超尺度变化的空间目标进行建模。提出一种遥感船舶的超尺度自导注意力网络(HSSGAN)识别框架,通过组连接的轻量级超尺度子空间模块捕获船舶的超尺度特征和尺度不变性,使用自导注意力网络逐步细化超尺度特征图,并在超尺度局部和全局语义之间建立长期依赖关系以增强类之间特征图的差异性。同时,通过忽略不相关信息及聚合相关特征以增强目标船舶的识别性。实验结果表明,与TP-FCN、CF-SDN和HSF-Net方法相比,HSSGAN方法具有更好的识别效果,F1-Score值为0.966 78。

关键词: 目标识别, 遥感图像, 卷积神经网络, 超尺度特征, 子空间模块, 组连接

Abstract: Due to the limited reception domain, the existing multi-scale Convolutional Neural Networks(CNN) often fail to model space targets with super-scale variation.In order to solve this problem, a Hyper-Scale Self-Guided Attention Networks(HSSGAN) recognition framework for remote sensing ships is proposed.The framework employs a lightweight super-scale subspace module connected by groups to capture the super-scale feature and scale invariance of the ship.Then the super-scale feature map is refined gradually by using the self-guided attention, and a long-term dependency relationship is established between the super-scale local and global semantics adaptively to enhance the difference of the feature maps between classes.In addition, irrelevant information is ignored while relevant features are aggregated, so the identifiability of the target ship can be enhanced.The experimental results show that the HSSGAN method exhibits improved recognition performance with the F1 value reaching 0.966 78.

Key words: target recognition, remote sensing images, Convolutional Neural Network(CNN), hyper-scale features, subspace module, group connection

中图分类号: