A context-aware transformer with weighted residual for efficient remote sensing target detection
Zhongyu Li, Yimin Shen, Xiaoping Jing, Rong Wang, Di Zhou, Yangyong Guo, Wei Yu, Fei Du · 2026
Read A context-aware transformer with weighted residual for efficient remote sensing target detection by Zhongyu Li, Yimin Shen, Xiaoping Jing, Rong Wang, Di Zhou, Yangyong Guo, Wei Yu, Fei Du (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/pmc-13219329.txt.
Description
Zhongyu Li and seven co-authors proposed in 2026 a detector for spotting targets in satellite images, tuned for disaster work. Standard models blur scales together and let class imbalance skew training. Their design pairs a context-aware transformer with a weighted residual pyramid that merges deep and shallow features. A learnable factor balances how much each layer contributes. Rotated boxes hug targets more tightly than plain rectangles.
A custom loss keeps any single target type from dominating the learning signal. On two public aerial benchmarks the model scored about 0.75 and 0.71 on the standard accuracy measure while staying fast enough for real-time use. Pictures from disaster scenes showed clean detections. The authors claim strong practical value for response mapping. The work joins a push for detectors that read context, not just shapes.
Catalog record
| CREATOR | Zhongyu Li, Yimin Shen, Xiaoping Jing, Rong Wang, Di Zhou, Yangyong Guo, Wei Yu, Fei Du |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 72 KB |
| VERSION | 1.0 |
| LICENSE | CC BY 4.0 |
| RIGHTS | CC BY |
| SOURCE | Europe PMC |