What question did this study set out to answer?

The aim is to enhance change detection in remote sensing imagery by integrating multi-scale and bi-temporal information effectively.

January 25, 2026Remote SensingOpen Access

MISA-Net: Multi-Scale Interaction and Supervised Attention Network for Remote-Sensing Image Change Detection

View Full Paper

Ask AI

Bookmark

Authors

This model demonstrates improved change detection in remote sensing images, suggesting effective multi-scale representation integration.

The aim is to enhance change detection in remote sensing imagery by integrating multi-scale and bi-temporal information effectively.
Developed the Multi-scale Interaction and Supervised Attention Network (MISANet).
Created a Progressive Multi-Scale Feature Fusion Module (PMFFM) for feature integration.
Introduced a Difference-guided Gated Attention Interaction (DGAI) module to improve interaction between bi-temporal features.
Implemented a Supervised Attention Decoder Module (SADM) for focused change region attention.
Used a deep supervision strategy to refine model training based on texture and semantic changes.
Achieved F1-scores of 91.19%, 82.25%, and 88.35% on LEVIR-CD, SYSU-CD, and GZ-CD respectively.
Improved performance by 0.50% F1 and 0.85% IoU on LEVIR-CD compared to BASNet.
Enhanced F1 and IoU by 2.13% and 3.02% on SYSU-CD.
Gained 1.28% F1 and 2.03% IoU on GZ-CD, demonstrating strong generalization abilities.

Yin et al. (2026) studied this question.

View Full Paper

Ask AI

Bookmark