Scholarlink

<p>Roads are essential infrastructures, and obtaining 3up-to-date road information is crucial for development. Optical satellite and aerial images serve as reliable and widely available data sources for road extraction. In this study, a multiscale geometric analysis (MGA) approach integrated with perceptual grouping methods was used to extract roads from high-resolution remote sensing images. A four-stage approach, each possessed distinct stapes: (1) the pulse-coupled neural network (PCNN) for deriving candidate road segments, (2) the curvelet transform (CT) for image decomposition and selective enhancement, (3) hysteresis thresholding for preliminary road structure extraction, and (4) tensor voting to infer and bridge gaps in the road network was used. Rigorous experimental investigations were conducted on IKONOS, GeoEye, and Google Earth images. The results demonstrate that the PCNN effectively identifies candidate road segments, the CT enhanced the image details, hysteresis thresholding extracted preliminary road structures, and the tensor voting successfully completed the detection and extraction of the road network. The approach achieved an average completeness and precision of over 0.87. F1 is 0.98 and IoU = 0.82, respectively. Compared to existing methods, this approach demonstrated broader applicability and effectiveness across diverse datasets and environments. It is a robust and efficient method for automatic extraction of roads from images containing complex background, associated with roads with intensity variation.</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

Multiscale geometric analysis-based automatic extraction of roads from high-resolution remote sensing images

  • 期刊论文

  • 发表日期：

    2026年2月25日

查看全文
  • Jiannong Cao,
  • Muleta Ebissa Feyissa,
  • Junjun Li

摘要

Roads are essential infrastructures, and obtaining 3up-to-date road information is crucial for development. Optical satellite and aerial images serve as reliable and widely available data sources for road extraction. In this study, a multiscale geometric analysis (MGA) approach integrated with perceptual grouping methods was used to extract roads from high-resolution remote sensing images. A four-stage approach, each possessed distinct stapes: (1) the pulse-coupled neural network (PCNN) for deriving candidate road segments, (2) the curvelet transform (CT) for image decomposition and selective enhancement, (3) hysteresis thresholding for preliminary road structure extraction, and (4) tensor voting to infer and bridge gaps in the road network was used. Rigorous experimental investigations were conducted on IKONOS, GeoEye, and Google Earth images. The results demonstrate that the PCNN effectively identifies candidate road segments, the CT enhanced the image details, hysteresis thresholding extracted preliminary road structures, and the tensor voting successfully completed the detection and extraction of the road network. The approach achieved an average completeness and precision of over 0.87. F1 is 0.98 and IoU = 0.82, respectively. Compared to existing methods, this approach demonstrated broader applicability and effectiveness across diverse datasets and environments. It is a robust and efficient method for automatic extraction of roads from images containing complex background, associated with roads with intensity variation.