Scholarlink

<p>Gravitational-wave observatories such as laser interferometric gravitational-wave observatories often encounter transient noise artifacts, or glitches, which obscure astrophysical signals and compromise detection accuracy. Effective glitch classification is vital to improving the sensitivity and reliability of these observatories. To develop a robust, interpretable, and generalizable deep learning framework capable of accurately classifying glitches in gravitational-wave data. The proposed Compression-based Hamiltonian Relevance Aware Capsule Network with Crisscross Moss Growth Optimization (C-HRACN-CMGO) introduces two central innovations. First, it embeds Hamiltonian dynamics into capsule-based modeling, enabling physically consistent representation of glitch evolution while preserving hierarchical spatial–temporal structures. Second, it integrates a compression-driven relevance mechanism, making the model lightweight and deployable in real-time observatories while enhancing interpretability through DeepLIFT. Multi-time window fusion with EfficientNetB7 strengthens spatial feature extraction, while CMGO fine-tunes network architecture, and regularization with label smoothing improves generalization. The framework achieved 98% classification accuracy, 99% precision, and 97.9% recall, outperforming existing methods. It is designed for real-time integration with the Gravity Spy platform, supporting online learning and adaptation. Beyond incremental improvements, the novelty lies in the synergistic design of physics-informed Hamiltonian capsule modeling and compression-aware interpretability, which ensures both high accuracy and real-time adaptability. This distinguishes the proposed model from conventional deep learning pipelines and supports scalable integration within the Gravity Spy platform.</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

An Effective Fusion and Deeplift-based Optimized Hamiltonian Relevance Aware Capsule Network for Glitch Classification in LIGO

  • 期刊论文

  • 发表日期：

    2025年11月24日

查看全文
  • Surbhi Agrawal

摘要

Gravitational-wave observatories such as laser interferometric gravitational-wave observatories often encounter transient noise artifacts, or glitches, which obscure astrophysical signals and compromise detection accuracy. Effective glitch classification is vital to improving the sensitivity and reliability of these observatories. To develop a robust, interpretable, and generalizable deep learning framework capable of accurately classifying glitches in gravitational-wave data. The proposed Compression-based Hamiltonian Relevance Aware Capsule Network with Crisscross Moss Growth Optimization (C-HRACN-CMGO) introduces two central innovations. First, it embeds Hamiltonian dynamics into capsule-based modeling, enabling physically consistent representation of glitch evolution while preserving hierarchical spatial–temporal structures. Second, it integrates a compression-driven relevance mechanism, making the model lightweight and deployable in real-time observatories while enhancing interpretability through DeepLIFT. Multi-time window fusion with EfficientNetB7 strengthens spatial feature extraction, while CMGO fine-tunes network architecture, and regularization with label smoothing improves generalization. The framework achieved 98% classification accuracy, 99% precision, and 97.9% recall, outperforming existing methods. It is designed for real-time integration with the Gravity Spy platform, supporting online learning and adaptation. Beyond incremental improvements, the novelty lies in the synergistic design of physics-informed Hamiltonian capsule modeling and compression-aware interpretability, which ensures both high accuracy and real-time adaptability. This distinguishes the proposed model from conventional deep learning pipelines and supports scalable integration within the Gravity Spy platform.