Scholarlink

<p>To address the challenge of Identifying traditional villages of diverse cultural types, this paper proposes an intelligent image classification method based on deep learning. We use five cultures—Hakka, Linchuan, Luling, Yuanzhou, and Yuzhang—as class labels and build a Traditional Villages in the Ganjiang River Basin dataset (TVITGRBOC) through data collection. After comparing several image classification models, we choose ConvNeXt and improve it. By introducing the HAT_SCConv module and a training strategy with hierarchical unfreezing and CutMix data augmentation, we enhance the model’s ability to extract fine-grained features and understand complex scenes. Experiments show that the improved ConvNeXt reaches about 86% on Accuracy, Precision, Recall, and F1_Score—about a 3% gain over the baseline. Compared with traditional manual methods, this approach greatly improves the efficiency and speed of Identifying traditional villages in the Ganjiang River Basin and provides an interpretable technical path for regional culture oriented digital preservation of traditional villages.</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

A deep learning-based method for identifying traditional villages across cultural types

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年12月19日

查看全文
  • Zhihao LI,
  • Dingfei LI,
  • Wenxuan Lai,
  • Zeyu Liu,
  • Qiliang Cai,
  • Chunqing Liu,
  • Cheng Zhang

摘要

To address the challenge of Identifying traditional villages of diverse cultural types, this paper proposes an intelligent image classification method based on deep learning. We use five cultures—Hakka, Linchuan, Luling, Yuanzhou, and Yuzhang—as class labels and build a Traditional Villages in the Ganjiang River Basin dataset (TVITGRBOC) through data collection. After comparing several image classification models, we choose ConvNeXt and improve it. By introducing the HAT_SCConv module and a training strategy with hierarchical unfreezing and CutMix data augmentation, we enhance the model’s ability to extract fine-grained features and understand complex scenes. Experiments show that the improved ConvNeXt reaches about 86% on Accuracy, Precision, Recall, and F1_Score—about a 3% gain over the baseline. Compared with traditional manual methods, this approach greatly improves the efficiency and speed of Identifying traditional villages in the Ganjiang River Basin and provides an interpretable technical path for regional culture oriented digital preservation of traditional villages.