PulseExploreJournal ClubResearchersJournals
Instagram
HomeJournal ClubExplore
Synapse
⌘+K
Synapse
March 21, 2026International Journal of Applied Earth Observation and GeoinformationOpen Access

MISNet: Multi-task interaction Siamese network for 3D point cloud semantic change detection

View Full Paper
Ask AI
Bookmark
Share

Authors

WZWenxiao ZhanWPWeiyue PengRCRuozhen Cheng

Discussion

Loading...

Member takes

Overview

Multi-task learning architecture improves semantic change detection in 3D point clouds, suggesting effective modeling techniques.

Key Points

  • The aim is to enhance 3D point cloud semantic change detection through a multi-task learning architecture.
  • Developed two datasets, HKSCD and UtrechtCD, utilizing photogrammetry and LiDAR point clouds.
  • Proposed the Multi-task Interaction Siamese Network (MISNet) architecture.
  • Implemented a multi-dimensional change encoding module to analyze temporal neighborhood relationships.
  • Introduced change-guided semantic refinement and semantic-awareness interaction modules for better feature representation.
  • MISNet achieves a mean Intersection over Union (mIoU) of 84.15% on HKSCD, 85.15% on UtrechtCD, and 89.58% on Urb3DCD-V2.
  • Outperformed existing methods by 2.21%, 1.43%, and 1.46% on respective datasets.

Cite This Study

Zhan et al. (2026) studied this question.

synapsesocial.com/papers/69be38216e48c4981c678496https://doi.org/10.1016/j.jag.2026.105246
View Full Paper
Ask AI
Bookmark
Share