Scholarlink

<p>Recent CNN–Transformer hybrids have advanced hyperspectral image (HSI) classification but often treat spectral and spatial features separately. To address this limitation, a spectral–spatial feature synergistic network (SSFSNet) is proposed, integrating three complementary modules for joint representation. The spectral feature marking (SFM) module extracts and compresses spectral information via convolution and attention. The spatial feature enhancement (SFE) module applies differential convolution and an multilayer perceptron to refine spatial structures and integrate context. The spatial channel synergistic attention (SCSA) module employs parallel convolutions and channel attention to capture multi-scale dependencies and strengthen spectral–spatial correlation. Experiments on three widely used public HSI datasets demonstrate that SSFSNet achieves superior classification performance compared to eight existing methods, with classification accuracies of 98.81±0.56%, 99.34±0.21%, and 99.40±0.18%, respectively. These results highlight the effectiveness and robustness of SSFSNet in complex classification scenarios. The source codes are available at: <a href="https://github.com/ZZP125/SSFSNet">https://github.com/ZZP125/SSFSNet</a>.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Enhancing hyperspectral image classification through spectral-spatial synergy: SSFSNet

  • 期刊论文

  • 发表日期：

    2026年1月8日

查看全文
  • Zaoping Zhong,
  • Chengbin Liang,
  • Ming Yang,
  • Deguang Wang

摘要

Recent CNN–Transformer hybrids have advanced hyperspectral image (HSI) classification but often treat spectral and spatial features separately. To address this limitation, a spectral–spatial feature synergistic network (SSFSNet) is proposed, integrating three complementary modules for joint representation. The spectral feature marking (SFM) module extracts and compresses spectral information via convolution and attention. The spatial feature enhancement (SFE) module applies differential convolution and an multilayer perceptron to refine spatial structures and integrate context. The spatial channel synergistic attention (SCSA) module employs parallel convolutions and channel attention to capture multi-scale dependencies and strengthen spectral–spatial correlation. Experiments on three widely used public HSI datasets demonstrate that SSFSNet achieves superior classification performance compared to eight existing methods, with classification accuracies of 98.81±0.56%, 99.34±0.21%, and 99.40±0.18%, respectively. These results highlight the effectiveness and robustness of SSFSNet in complex classification scenarios. The source codes are available at: https://github.com/ZZP125/SSFSNet.