【摘要】 针对复杂太空环境中非合作目标位姿估计面临的背景干扰和尺度变化挑战，提出一种基于背景解耦的多尺度空间目标位姿估计方法．该方法采用双路并行架构，同时处理目标图像和纯背景图像，通过特征对比学习机制实现目标特征与环境噪声的有效分离．网络架构设计上，结合EfficientNet骨干网络的深度可分离卷积特征提取能力与BiFPN模块的双向跨尺度特征融合优势，构建了高效的多尺度特征表征体系．仿真实验表明：所提出的方法在SwissCube数据集上实现位置误差低至0.107 m，姿态误差低至2.164°，性能指标显著优于现有方法．更多还原

【Abstract】 To address the challenges of background interference and scale variation in non-cooperative target pose estimation in complex space environments,a multi-scale keypoint detection network based on background decoupling was proposed.The method employed a dual-path parallel architecture that simultaneously processed both target images and pure background images,achieving effective separation of target features from environmental noise through a feature contrastive learning mechanism. In terms of network architecture design,an efficient multi-scale feature representation system was constructed by combining the depthwise separable convolution feature extraction capability of the EfficientNet backbone network with the bidirectional cross-scale feature fusion advantages of the BiFPN(bidirectional feature pyramid network) module. Simulation experiments show that the proposed method achieves a position error as low as 0.107 m,and an attitude error of merely 2.164° on the SwissCube dataset,with performance metrics significantly surpassing existing approaches.更多还原