Scholarlink

<p>Oil painting classification has become a significant factor in the fields of art history and computer vision, although several issues persist due to the complexity and diversity of artistic styles. ResNet50-based SD-ResNet50 is a deep learning model introduced in this paper that features a Squeeze and Excitation module to facilitate inter-channel relationships and a Dilated Convolution-Transformer module to fetch local and global features. A multilayer perceptron also helps in improving the classification result. The Pandora dataset, which includes 12 different painting styles, was used to test SD-ResNet50, and the obtained results were quite good: accuracy (92.55%), precision (93.14%), recall (90.72%), the F1-score (91.84%), and mAP (91.42%), among others. The use of class activation maps indicated that it was possible to localize the most stylistic features of the system, which represents a significant step forward in the field of machine art analysis.</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

Deep learning-driven oil painting style recognition and classification: exploration and innovation

  • 期刊论文

  • 发表日期：

    2025年9月5日

查看全文
  • Jingjing Jin

摘要

Oil painting classification has become a significant factor in the fields of art history and computer vision, although several issues persist due to the complexity and diversity of artistic styles. ResNet50-based SD-ResNet50 is a deep learning model introduced in this paper that features a Squeeze and Excitation module to facilitate inter-channel relationships and a Dilated Convolution-Transformer module to fetch local and global features. A multilayer perceptron also helps in improving the classification result. The Pandora dataset, which includes 12 different painting styles, was used to test SD-ResNet50, and the obtained results were quite good: accuracy (92.55%), precision (93.14%), recall (90.72%), the F1-score (91.84%), and mAP (91.42%), among others. The use of class activation maps indicated that it was possible to localize the most stylistic features of the system, which represents a significant step forward in the field of machine art analysis.