Style Alignment-Based Dynamic Observation Method for UAV-View Geo-Localization Article Swipe
YOU?
·
· 2023
· Open Access
·
· DOI: https://doi.org/10.1109/tgrs.2023.3337383
The task of UAV-view geo-localization is to estimate the localization of a\nquery satellite/drone image by matching it against a reference dataset\nconsisting of drone/satellite images. Though tremendous strides have been made\nin feature alignment between satellite and drone views, vast differences in\nboth inter and intra-class due to changes in viewpoint, altitude, and lighting\nremain a huge challenge. In this paper, a style alignment based dynamic\nobservation method for UAV-view geo-localization is proposed to meet the above\nchallenges from two perspectives: visual style transformation and surrounding\nnoise control. Specifically, we introduce a style alignment strategy to\ntransfrom the diverse visual style of drone-view images into a unified\nsatellite images visual style. Then a dynamic observation module is designed to\nevaluate the spatial distribution of images by mimicking human observation\nhabits. It is featured by the hierarchical attention block (HAB) with a\ndual-square-ring stream structure, to reduce surrounding noise and geographical\ndeformation. In addition, we propose a deconstruction loss to push away\nfeatures of different geo-tags and squeeze knowledge from unmatched images by\ncorrelation calculation. The experimental results demonstrate the\nstate-of-the-art performance of our model on benchmarked datasets. In\nparticular, when compared to the prior art on University-1652, our results\nsurpass the best of them (FSRA), while only requiring 2x fewer parameters. Code\nwill be released at https://github.com/Xcco1/SA\\_DOM\n
Related Topics To Compare & Contrast
- Type
- article
- Language
- en
- Landing Page
- https://doi.org/10.1109/tgrs.2023.3337383
- OA Status
- green
- Cited By
- 14
- References
- 48
- Related Works
- 10
- OpenAlex ID
- https://openalex.org/W4389104778