融合多层级特征的遥感图像地面弱小目标检测
DOI:
CSTR:
作者:
作者单位:

作者简介:

通讯作者:

中图分类号:

TP751. 1 TP753 TH701

基金项目:

中央高校基本科研业务费资助(NJ2020021)、中央高校基本科研业务费资助(NT2020022)、国家自然科学基金(61901504)项目资助 ,国家自然科学基金资助(61705104);中央高校基本科研业务费资助(NJ2020021);江苏省自然科学基金资助(BK20170804)


Multi-level feature fusion based dim small ground target detection in remote sensing images
Author:
Affiliation:

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:

    为解决遥感图像地面弱小目标检测中弱小目标信息量少、信息真假混杂的难题,本文提出一种融合多层级特征的遥感 图像地面弱小目标检测算法 CC-YOLO。 该算法首先利用深度卷积神经网络逐级对目标图像进行特征提取,得到高低层特征空 间金字塔图;然后,对空间金字塔图进行跨层级通道特征融合,结合新增的位置注意力机制 CA,分别沿两个空间方向聚合特征, 保留弱小目标精确的位置信息;最后,在聚合后生成的双支路特征图上进行端到端的目标检测,联合多通道检测信息输出检测 结果。 为解决算法实验中图像数据匮乏的问题,构建了遥感图像地面弱小目标数据集 GDSTD。 实验结果表明,算法 AP0. 5 ∶0. 95 达到 42. 3% ,AP0. 5 达到 94. 6% ,检测速率 FPS 达到 58. 8 帧/ s,具有一定的鲁棒性和实时性。

    Abstract:

    The detection of dim small ground targets in remote sensing images has problems of less target information and mixed information. To address these issues, a detection algorithm based on the multi-level feature fusion is proposed in this article, which is named as CC-YOLO. Firstly, the deep convolution neural network is used to extract features of the target image step by step, and the high-level and low-level feature spatial pyramid is obtained. Then, the cross-level channel feature fusion is implemented on the spatial pyramid, and features are aggregated along two spatial directions. The newly added CA is combined to retain the accurate location information of dim small targets. Finally, the end-to-end target detection method is implemented on the dual feature map generated after aggregation. And the detection results are output by combining multi-channel detection information. To solve the problem of lacking image data in algorithm experiment, this article establishes the ground-based dim small target dataset ( GDSTD) of remote sensing image. Experimental results show that the proposed algorithm achieves 42. 3% at AP0. 5 ∶0. 95 and 94. 6% at AP0. 5 , and the detection rate FPS reaches 58. 8 frames / s, which has certain robustness and real-time performance.

    参考文献
    相似文献
    引证文献
引用本文

闫钧华,张 琨,施天俊,朱桂熠,刘 勇,张 寅.融合多层级特征的遥感图像地面弱小目标检测[J].仪器仪表学报,2022,43(3):221-229

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: 2023-02-06
  • 出版日期:
文章二维码