航空学报 > 2025, Vol. 46 Issue (23): 631961-631961   doi: 10.7527/S1000-6893.2025.31961

干扰环境下无人机多源感知专栏

基于图像翻译的飞行器红外/卫星异源快速匹配定位方法

唐彬, 杨小冈(), 卢瑞涛, 张震宇, 宿爽   

  1. 火箭军工程大学 导弹工程学院,西安 710025
  • 收稿日期:2025-03-11 修回日期:2025-03-26 接受日期:2025-05-22 出版日期:2025-06-10 发布日期:2025-06-06
  • 通讯作者: 杨小冈 E-mail:doctoryyxg@163.com
  • 基金资助:
    陕西省重点研发计划重点项目(2024CY2-GJHX-42)

Aircraft infrared/satellite heterogenous fast matching localization method based on image translation

Bin TANG, Xiaogang YANG(), Ruitao LU, Zhenyu ZHANG, Shuang SU   

  1. College of Missile Engineering,Rocket Force University of Engineering,Xi’an 710025,China
  • Received:2025-03-11 Revised:2025-03-26 Accepted:2025-05-22 Online:2025-06-10 Published:2025-06-06
  • Contact: Xiaogang YANG E-mail:doctoryyxg@163.com
  • Supported by:
    Key Research and Development Program of Shaanxi Province(2024CY2-GJHX-42)

摘要:

在全球导航卫星系统拒止环境下,飞行器视觉导航面临夜间异源图像差异显著和大视场旋转导致特征匹配失效的双重挑战。为此,提出了一种基于图像翻译飞行器红外/卫星异源快速匹配定位方法,旨在通过图像翻译与预旋转匹配策略提升飞行器夜间匹配定位的性能。首先,构建了HC_CycleGAN模型,通过结合Huber损失与空间注意力机制实现卫星影像至红外域的跨域转换,解决异源图像的域对齐问题。其次,设计了FastPoint快速特征点提取网络,基于设计的深度可变卷积层,结合残差学习机制构建了R-DVM计算单元,以提高模型计算效率及训练稳定性。最后提出基于L-LightGlue动态自适应匹配算法的旋转匹配方法,结合几何不变预旋转匹配策略进行旋转匹配校正,利用图像间变换关系确定飞行器在卫星图像中的位置,完成视觉定位。实验结果表明,相比于其他匹配定位方法,所提方法在提高匹配效率的同时,能够显著提升大视场旋转条件下的红外/卫星异源图像匹配定位的精度。

关键词: 异源图像, 图像翻译, 特征提取, 旋转匹配, 飞行器定位

Abstract:

In Global Navigation Satellite System-denied environments, aircraft visual navigation faces dual challenges posed by significant nighttime heterogenous image discrepancies and feature matching failures caused by large field-of-view rotations. To address these issues, this paper proposes an infrared/satellite cross-domain fast matching localization method for aircraft based on image translation, aiming to enhance nighttime matching and localization performance through image translation and rotational matching strategies. First, we construct HC_CycleGAN, a cross-domain translation model that leverages the integration of Huber loss and spatial attention mechanisms to align satellite and infrared domains. Second, we develop FastPoint, a rapid feature extraction network that integrates a deep variable convolutional layer with residual learning mechanisms to establish an R-DVM computational unit, enhancing both computational efficiency and training stability. Finally, a rotational matching method based on the L-LightGlue dynamic adaptive matching algorithm is proposed. This method combines a geometry-invariant pre-rotation matching strategy for rotational matching correction to determine the aircraft’s position in satellite image based on the inter-image transformation relationships, thereby accomplishing visual localization. Experimental results demonstrate that, compared to existing matching and localization methods, the proposed approach not only improves matching efficiency but also significantly reduces heterogenous image matching errors under rotational conditions.

Key words: heterogenous images, image translation, feature extraction, rotation matching, aircraft localization

中图分类号: