Scholarlink

<p>Effective feature extraction is a key challenge in ground-based cloud classification. However, existing algorithms often rely on hand-crafted features, demanding precise manual labeling and resulting in limited classification accuracy. While deep learning offers promise, capturing the crucial salient features within the complex texture of cloud images remains difficult. To address these issues, we introduce a weakly supervised ground-based cloud classification approach (WS-GCCA). WS-GCCA employs a two-branch architecture, using coarse-grained and fine-grained deep networks to extract complementary global and local features. The method was validated on a ground-based cloud classification database with 11 cloud types. Experiments show that WS-GCCA achieves a classification accuracy of 98.58%, significantly outperforming 10 state-of-the-art supervised learning algorithms.</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

Ground-based cloud classification algorithm based on weakly supervised attention learning

  • 期刊论文

  • 发表日期：

    2025年7月15日

查看全文
  • Leyuan Xing,
  • Zhonghua Sun,
  • Kebin Jia,
  • Jinchao Feng

摘要

Effective feature extraction is a key challenge in ground-based cloud classification. However, existing algorithms often rely on hand-crafted features, demanding precise manual labeling and resulting in limited classification accuracy. While deep learning offers promise, capturing the crucial salient features within the complex texture of cloud images remains difficult. To address these issues, we introduce a weakly supervised ground-based cloud classification approach (WS-GCCA). WS-GCCA employs a two-branch architecture, using coarse-grained and fine-grained deep networks to extract complementary global and local features. The method was validated on a ground-based cloud classification database with 11 cloud types. Experiments show that WS-GCCA achieves a classification accuracy of 98.58%, significantly outperforming 10 state-of-the-art supervised learning algorithms.