Scholarlink

<p>Cloud detection plays a significant role in various practices, such as weather forecasting, climate studies, etc. This paper presents an integrated approach called cloud-WAVECAP for ground-based cloud detection using a combination of wavelet and capsule networks. The proposed approach utilizes the wavelet’s multiscale analysis capability to detect significant cloud features at different resolutions. Meanwhile, the inherent capability of dynamic routing by the capsule network increases the model’s ability to capture hierarchical structures and spatial relationships within the clouds. The cloud-WAVECAP model applies two-level wavelet decomposition, followed by convolutional layers and the capsule network. This architecture integrates preprocessing, wavelet layers, and capsule layers to capture both low- and high-level features for accurate cloud classification. It excels in identifying different cloud types, which is vital for meteorological analysis. Cloud-WAVECAP is assessed using several metrics and outperforms Inception V3, VGGNet, Resnet50, EfficientNet-B7, achieving 98.42% precision, 98.48% recall, and 99.12% accuracy. Additionally, the model’s efficiency, measured by Floating Point Operations (FLOPs), is competitive, resulting in 1.5426 GIGA FLOPS compared to other methods.</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

Cloud-WAVECAP: Ground-based cloud types detection with an efficient wavelet-capsule approach

  • 期刊论文

  • 发表日期：

    2025年1月23日

查看全文
  • Sanjukta Mishra,
  • Samarjit Kar,
  • Parag Kumar Guhathakurta

摘要

Cloud detection plays a significant role in various practices, such as weather forecasting, climate studies, etc. This paper presents an integrated approach called cloud-WAVECAP for ground-based cloud detection using a combination of wavelet and capsule networks. The proposed approach utilizes the wavelet’s multiscale analysis capability to detect significant cloud features at different resolutions. Meanwhile, the inherent capability of dynamic routing by the capsule network increases the model’s ability to capture hierarchical structures and spatial relationships within the clouds. The cloud-WAVECAP model applies two-level wavelet decomposition, followed by convolutional layers and the capsule network. This architecture integrates preprocessing, wavelet layers, and capsule layers to capture both low- and high-level features for accurate cloud classification. It excels in identifying different cloud types, which is vital for meteorological analysis. Cloud-WAVECAP is assessed using several metrics and outperforms Inception V3, VGGNet, Resnet50, EfficientNet-B7, achieving 98.42% precision, 98.48% recall, and 99.12% accuracy. Additionally, the model’s efficiency, measured by Floating Point Operations (FLOPs), is competitive, resulting in 1.5426 GIGA FLOPS compared to other methods.