Image retrieval has emerged as a robust complementary approach for accurate Unmanned Aerial Vehicle (UAV) self-positioning. However, current methodologies often prioritize object localization and struggle with the unique demands of precise UAV self-positioning, particularly fine-grained spatial discrimination and robustness against dynamic scene variations. To overcome these limitations, we propose the Context-Enhanced method for precise UAV Self-Positioning (CEUSP). CEUSP incorporates a Dynamic Sampling Strategy for efficient negative sample selection, alongside a distinct feature learning pipeline. This pipeline features a Rubik’s Cube Attention module, which, inspired by the rotational mechanics of a Rubik’s Cube, effectively models interdimensional feature interactions, complemented by a Context-Aware Channel Integration module to enhance representation and discriminability. Extensive experimental validation demonstrates CEUSP’s efficacy, achieving significant improvements in both feature representation and UAV self-positioning accuracy, especially in complex urban environments. Our approach establishes state-of-the-art performance on the challenging DenseUAV dataset, specifically designed for dense urban contexts, and achieves competitive results on the widely recognized University-1652 benchmark. Implementation code and experimental results are available at https://github.com/eksnew/CEUSP .