Scholarlink

This study presents an approach to isolines extraction in topographic maps through the integration of deep learning techniques with automated training data generation. A geologic map image generator based on Perlin Noise was developed to augment existing datasets, addressing the challenge of limited annotated data in geospatial informatics. The generated synthetic data improved the performance of U-Net models in semantic segmentation tasks. Experimental evaluations revealed a 6% increase in the Dice coefficient and a 54% rise in precision compared to baseline models. These results highlight the effectiveness of the proposed method in enhancing the generalizability of isolines extraction systems, making it a valuable tool for automating geospatial data processing and analysis.

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

U-Net Models Enhanced by Generated Training Data for Automatic Isolines Extraction

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Vladislav Kurochkin,
  • Yury Karyakin,
  • Irina Donkova

摘要

This study presents an approach to isolines extraction in topographic maps through the integration of deep learning techniques with automated training data generation. A geologic map image generator based on Perlin Noise was developed to augment existing datasets, addressing the challenge of limited annotated data in geospatial informatics. The generated synthetic data improved the performance of U-Net models in semantic segmentation tasks. Experimental evaluations revealed a 6% increase in the Dice coefficient and a 54% rise in precision compared to baseline models. These results highlight the effectiveness of the proposed method in enhancing the generalizability of isolines extraction systems, making it a valuable tool for automating geospatial data processing and analysis.