Scholarlink

<p>Automation of scene classification within artistic domains has become more prevalent due to the advent of deep learning. This study presented a new framework for classifying dance scenes using an adapted Squeeze-and-Excitation Network (SENet) and optimized hyperparameters with Advanced Beluga Whale Optimization Algorithm (ABWOA). The SENet architecture allows for redistributing features by adaptively focusing on the most discriminative visual features (e.g., posture, costume), and ABWOA efficiently optimizes hyper-parameters improve convergence and generalization. The framework has been evaluated on the Dance5K dataset that significantly outperformed the state-of-the-art models with an accuracy value of 92.6% and an F1-score value of 96.4%. Comprehensive experiments, including statistical significance tests, and cross validation validates the robustness of the framework. These results illustrate the potential of blending attention-based CNNs and bio-inspired optimization techniques within fine-grained visual categorization of cultural and artistic domains.</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

Scene classification in dance art using deep learning with squeeze and excitation networks and the Advanced Beluga Whale Optimization Algorithm

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年4月6日

查看全文
  • Yude Dong,
  • Linfang Sun

摘要

Automation of scene classification within artistic domains has become more prevalent due to the advent of deep learning. This study presented a new framework for classifying dance scenes using an adapted Squeeze-and-Excitation Network (SENet) and optimized hyperparameters with Advanced Beluga Whale Optimization Algorithm (ABWOA). The SENet architecture allows for redistributing features by adaptively focusing on the most discriminative visual features (e.g., posture, costume), and ABWOA efficiently optimizes hyper-parameters improve convergence and generalization. The framework has been evaluated on the Dance5K dataset that significantly outperformed the state-of-the-art models with an accuracy value of 92.6% and an F1-score value of 96.4%. Comprehensive experiments, including statistical significance tests, and cross validation validates the robustness of the framework. These results illustrate the potential of blending attention-based CNNs and bio-inspired optimization techniques within fine-grained visual categorization of cultural and artistic domains.