Spatial–Spectral ConvNeXt for Hyperspectral Image Classification
Yimin Zhu, Kexin Yuan, W. Zhong, Linlin Xu
Abstract
Hyperspectral image (HSI) classification is a difficult task due to the heterogeneous spatial-spectral information, high-dimensiontality and noise effect in HSI. Lately, an enhanced convolutional approach, i.e., ConvNeXt, demonstrates stronger feature representation capability than the popular vision transformer (ViT) approaches. This paper presents a spatial-spectral ConvNeXt approach, called SS-ConvNeXt, for hyperspectral classification. To better learn the spatial and spectral information in HSI, the Spatial-ConvNeXt (Spa-cv) block, Spectral-ConvNeXt (Spe-cv) block and Spectral Projection Module (SPM) are respectively designed. The depthwise and pointwise convolutions are adopted to reduce the model size and prevent vanishing gradient. The proposed model is evaluated against 14 other state-of-the-art (SOTA) methods on 4 different HSI datasets. Moreover, extensive ablation studies are conducted to investigate the roles of building blocks in the proposed model. The results demonstrate that the proposed method not only can achieve high classification accuracy but also can better preserve class boundaries and reduce within-class noise. The codes of this work will be publicly available on Github.