Scholarlink

To enhance classification performance and reduce annotation costs simultaneously, this paper introduces a dual-branch attention mechanism network (DBDAA) that integrates deep learning and active learning methods. In DBDAA, spectral and spatial features are extracted from hyperspectral images using two branches. To further refine and optimize the extracted feature maps, the network incorporates channel attention and spatial attention mechanisms. Due to the severe shortage of labeled samples in hyperspectral images, an uncertainty-based active learning approach is also introduced.

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

Hyperspectral Image Classification with Double-Branch Convolutional Neural Network and Active Learning

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Ruirui Zhang,
  • Liwei Chen,
  • Beiming Li,
  • Tong Wang,
  • Shan Gao,
  • Min Ouyang

摘要

To enhance classification performance and reduce annotation costs simultaneously, this paper introduces a dual-branch attention mechanism network (DBDAA) that integrates deep learning and active learning methods. In DBDAA, spectral and spatial features are extracted from hyperspectral images using two branches. To further refine and optimize the extracted feature maps, the network incorporates channel attention and spatial attention mechanisms. Due to the severe shortage of labeled samples in hyperspectral images, an uncertainty-based active learning approach is also introduced.