Scholarlink

<p>Detecting mural deterioration is of great significance for the preventive conservation, scientific restoration, and digital preservation of ancient murals. To this end, we propose a novel mural deterioration detection model based on Adaptive Feature Selection and Category-Aware Cascaded Decision (AFSCACD). This model first employs a Multi-view Feature Extraction (MFE) strategy to capture multiple visual features of murals, thereby comprehensively representing deterioration characteristics. Then, an Adaptive Feature Selection (AFS) strategy dynamically filters and matches features, highlighting discriminative representations while suppressing redundant information. Finally, the Category-Aware Cascaded Decision (CACD) strategy implements cascaded decisions according to the classification difficulty of deterioration cases, effectively improving the detection precision of hard samples and small-scale targets. Experimental results demonstrate that AFSCACD achieves superior performance over mainstream models, with an F1-score of 0.9370 exceeding the baseline by at least 3.55%, thus providing effective technical support for mural conservation.</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

Detecting ancient mural deterioration based on adaptive feature selection and category-aware cascaded decision

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年7月9日

查看全文
  • Shouqiang Sun,
  • Qingqing Li,
  • Xingshuai Ma,
  • Xinyu Yuan,
  • Yuhan Liu,
  • Jiarui Han

摘要

Detecting mural deterioration is of great significance for the preventive conservation, scientific restoration, and digital preservation of ancient murals. To this end, we propose a novel mural deterioration detection model based on Adaptive Feature Selection and Category-Aware Cascaded Decision (AFSCACD). This model first employs a Multi-view Feature Extraction (MFE) strategy to capture multiple visual features of murals, thereby comprehensively representing deterioration characteristics. Then, an Adaptive Feature Selection (AFS) strategy dynamically filters and matches features, highlighting discriminative representations while suppressing redundant information. Finally, the Category-Aware Cascaded Decision (CACD) strategy implements cascaded decisions according to the classification difficulty of deterioration cases, effectively improving the detection precision of hard samples and small-scale targets. Experimental results demonstrate that AFSCACD achieves superior performance over mainstream models, with an F1-score of 0.9370 exceeding the baseline by at least 3.55%, thus providing effective technical support for mural conservation.