.png){kind=icon}
Return
Nonparameter Attention-Guided Deformable Convolution Network for Hyperspectral Image Classification
.png){kind=icon}
PRE
AI
DOI：10.1109/LGRS.2026.3664312.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Convolutional neural networks (CNNs) have gained widespread use in hyperspectral classification tasks with their robust feature extraction capabilities. However, when dealing with high-dimensional hyperspectral data, the fixed kernel shape in the convolution operation is unable to dynamically adjust its sampling position to adapt to the intricate data structure. To alleviate this problem, we proposed a nonparameter attention-guided deformable convolution (NPA-DConv) as a plug-and-play feature extraction module, which improves deformable convolution with nonparameter attention (NPA) that does not require learnable parameters. More specifically, the NPA is designed as a representation of spatial–spectral feature importance, which is used to calculate the offset of the kernel deformation, enabling precise positioning of the sampling points. In this way, conventional convolution is replaced by the NPA-DConv to achieve dynamic adjustment of the kernel shape. On this basis, a hyperspectral classification framework, nonparameter attention-guided deformable convolution network (NPADCN), is tailored for the joint extraction of spatial–spectral discriminant features. Experimental results conducted on three public hyperspectral datasets demonstrate the superior performance of the proposed methods. The source code of the proposed NPADCN is available at https://github.com/wybable/NPADCN/
Keywords:
Attention mechanisms
convolutional neural networks (CNNs)
deformable convolution
hyperspectral image (HSI) classification
spatial–spectral feature
Journal
I
IF:
4.4
Papers:
754
Citations:
0
Organization
View more organizations
Cited Papers
.png){kind=icon}
ShareSaveShareSaveShareSaveShareSave View more