Synapse
⌘+K
Synapse
PulseExploreClubsResearchersJournals
Instagram
HomeClubsExplore
May 17, 2026Remote SensingOpen Access

Positive-Guided Local Supervision for Robust Road Extraction from Remote Sensing Imagery

View Full Paper
Ask AI
Bookmark
Share

Authors

HHHao HeSWShuyang WangLHLei Huang

Discussion

Loading...

Member takes

Overview

Randomized trial shows improved road extraction accuracy in remote sensing imagery, indicating enhanced performance against label noise.

Key Points

  • This research aims to develop a training strategy that improves road extraction accuracy in remote sensing imagery despite label noise.
  • Proposes Positive-guided Local Supervision (PLS) strategy to enhance road extraction accuracy.
  • Combines global context from dense networks with local supervision focusing on reliable annotations.
  • Evaluated on two datasets: DeepGlobe and a new dataset, China Four Provinces (CH4P).
  • PLS strategy achieved absolute IoU improvements of 0.127 on DeepGlobe and 0.104 on CH4P over baseline models.
  • Demonstrated superiority in handling underlabeling across datasets under both original and refined annotations.
  • Effectively retrieved omitted road segments without additional computational burden.

Cite This Study

He et al. (2026) studied this question.

synapsesocial.com/papers/6a095c3f7880e6d24efe2482https://doi.org/10.3390/rs18101589
View Full Paper
Ask AI
Bookmark
Share