.png){kind=icon}
.png){kind=icon}
Return
Spatial-Frequency Joint Learning Mamba for Hyperspectral Image Classification
.png){kind=icon}
PRE
AI
DOI：10.1109/LGRS.2026.3652369.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Hyperspectral image (HSI) classification, as a crucial research direction in remote sensing, aims to achieve pixel-wise land cover classification by leveraging the abundant spectral and spatial information of HSI. However, existing approaches direct their primary attention to spectral–spatial feature analysis while neglecting to explore latent properties in the frequency domain. To solve this issue, we put forward a spatial-frequency joint learning Mamba (SFMamba) framework for HSI classification. Its core component is the wavelet Mamba block (WMB), which adopts a dual-branch design with spatial and frequency branches. In the spatial branch, we utilize convolutional layers for spatial domain feature extraction. In the frequency branch, we use wavelet transform to decompose input features into a low-frequency component together with three high-frequency components. The low-frequency component is processed by a convolutional block, while the high-frequency components are modeled by a Mamba block, achieving effective learning of both local and global representations in the frequency domain. Experimental results on two public datasets show that SFMamba outperforms several existing classification methods, demonstrating its effectiveness. Specifically, SFMamba achieves overall accuracies of 98.92% on the ZY1-02D Huanghekou (ZYHHK) dataset and 99.03% on the WHU-Hi-LongKou (LongKou) dataset. Our code is available at: https://github.com/zhe-meng/SFMamba
Keywords:
Convolutional neural network (CNN)
frequency
hyperspectral image (HSI) classification
mamba
Journal
I
IF:
4.4
Papers:
572
Citations:
0