Adaptive pixel attention network for hyperspectral image classification
Yuefeng Zhao, Chengmin Zai, Nannan Hu, Lu Shi, Xue Zhou, Jingqi Sun · 2024
Read Adaptive pixel attention network for hyperspectral image classification by Yuefeng Zhao, Chengmin Zai, Nannan Hu, Lu Shi, Xue Zhou, Jingqi Sun (2024) 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-11585658.txt.
Description
Yuefeng Zhao and five co-authors build a neural network that sorts land cover from the many colors of hyperspectral images. Their 2024 paper tackles a flaw in the usual approach: fixed convolution kernels treat a patch of pixels rigidly, so the links between neighboring pixels go unlearned and accuracy suffers. Their network instead pays adaptive attention to the pixels themselves.
Three modules do the work. One enriches the spectral and spatial signal in the input cubes. The core attention step compares pixels by both distance and angle, learning which neighbors matter at several scales. A third module mixes features across layers so telling details are not lost between levels. On four standard data sets the network beat current classification models on accuracy, and it ran faster than rival three-dimensional designs. The authors aim the method at remote sensing tasks where fine spectral detail decides the label.
Catalog record
| CREATOR | Yuefeng Zhao, Chengmin Zai, Nannan Hu, Lu Shi, Xue Zhou, Jingqi Sun |
|---|---|
| YEAR | 2024 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 124 KB |
| VERSION | 1.0 |
| LICENSE | CC BY 4.0 |
| RIGHTS | CC BY |
| SOURCE | Europe PMC |