Scholarlink

<p>Rooftop extraction from remote sensing imagery supports various socio-economic applications but faces challenges due to the diversity and complexity of the building rooftops. This work introduces a deep learning model, namely the Intermediate Supervision Model with Asymmetric Pyramid Non-Local Blocks (ISAPNBNet), for facilitating high-precision building extraction. The proposed model provides robust segmentation capabilities by capturing long-range dependencies within the image data. In order to address the lack of Indian building extraction imagery, a dataset named C3-Hosur-MX (1.12&#xa0;m resolution) has been prepared from the study area of Hosur, Tamil Nadu, India. The performance of the proposed model is compared in key metrics with that of the UNet, SegNet, UNetVasyPPD, MRDCN and ISNET both qualitatively and quantitatively, using the publically available Massachusetts building dataset and C3-Hosur-MX dataset. The outcomes suggest that the proposed model can be used for extracting buildings in various environments.</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

Semantic segmentation model with asymmetric pyramid non-local blocks for building extraction from satellite imagery

  • 期刊论文

  • 发表日期：

    2026年6月2日

查看全文
  • Sam V. George,
  • R. Avudaiammal,
  • J. Martin Leo Manickam,
  • Vandita Srivastava

摘要

Rooftop extraction from remote sensing imagery supports various socio-economic applications but faces challenges due to the diversity and complexity of the building rooftops. This work introduces a deep learning model, namely the Intermediate Supervision Model with Asymmetric Pyramid Non-Local Blocks (ISAPNBNet), for facilitating high-precision building extraction. The proposed model provides robust segmentation capabilities by capturing long-range dependencies within the image data. In order to address the lack of Indian building extraction imagery, a dataset named C3-Hosur-MX (1.12 m resolution) has been prepared from the study area of Hosur, Tamil Nadu, India. The performance of the proposed model is compared in key metrics with that of the UNet, SegNet, UNetVasyPPD, MRDCN and ISNET both qualitatively and quantitatively, using the publically available Massachusetts building dataset and C3-Hosur-MX dataset. The outcomes suggest that the proposed model can be used for extracting buildings in various environments.