DHChange: learning under source and supervision heterogeneity for unified remote sensing change detection

Unified remote sensing change detection seeks to handle heterogeneous datasets and detection tasks within a single model. Existing approaches mainly focus on standardizing task inputs and outputs, while paying limited attention to two challenges that arise during joint learning: negative transfer among data sources with different visual distributions and insufficient knowledge exchange between binary change detection (BCD) and semantic change detection (SCD), which provide supervision at different semantic granularities. To address these challenges, we propose DHChange, a unified framework for learning under source and supervision heterogeneity. A factorized task–role query design first provides a shared instruction-conditioned interface for change localization and bi-temporal content interpretation. To mitigate cross-source representation conflicts, DHChange introduces shared–private prototype collaboration, in which shared prototypes encode transferable changed and unchanged patterns, while source-private prototypes preserve complementary source-dependent characteristics and are retrieved through a source-supervised adaptive router. To facilitate collaboration between BCD and SCD supervision, a bidirectional cross-granularity interaction module uses category-agnostic localization responses to enhance bi-temporal content representations and subsequently exploits their content differences to refine change localization. For semantic prediction, pixel-to-text matching provides a parameter-shared interface that is not tied to a fixed-dimensional linear classifier. Experiments on three BCD benchmarks and SECOND show improvements over common-encoder comparison variants of task-specific and unified methods, with controlled analyses supporting source-aware collaboration and complementary BCD/SCD supervision.

Authors

Institutions

Publication Details

Journal
Complex & Intelligent Systems
Published
2026-11-15
DOI
https://doi.org/10.1007/s40747-026-02582-z
Primary Topic
Remote-Sensing Image Classification
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

DHChange: learning under source and supervision heterogeneity for unified remote sensing change detection

Xingheng Wang, Yinuo Zhu
Complex & Intelligent Systems
Remote-Sensing Image Classification
article

DHChange: learning under source and supervision heterogeneity for unified remote sensing change detection

Xingheng Wang, Yinuo Zhu
article en

Abstract

Unified remote sensing change detection seeks to handle heterogeneous datasets and detection tasks within a single model. Existing approaches mainly focus on standardizing task inputs and outputs, while paying limited attention to two challenges that arise during joint learning: negative transfer among data sources with different visual distributions and insufficient knowledge exchange between binary change detection (BCD) and semantic change detection (SCD), which provide supervision at different semantic granularities. To address these challenges, we propose DHChange, a unified framework for learning under source and supervision heterogeneity. A factorized task–role query design first provides a shared instruction-conditioned interface for change localization and bi-temporal content interpretation. To mitigate cross-source representation conflicts, DHChange introduces shared–private prototype collaboration, in which shared prototypes encode transferable changed and unchanged patterns, while source-private prototypes preserve complementary source-dependent characteristics and are retrieved through a source-supervised adaptive router. To facilitate collaboration between BCD and SCD supervision, a bidirectional cross-granularity interaction module uses category-agnostic localization responses to enhance bi-temporal content representations and subsequently exploits their content differences to refine change localization. For semantic prediction, pixel-to-text matching provides a parameter-shared interface that is not tied to a fixed-dimensional linear classifier. Experiments on three BCD benchmarks and SECOND show improvements over common-encoder comparison variants of task-specific and unified methods, with controlled analyses supporting source-aware collaboration and complementary BCD/SCD supervision.

Complex & Intelligent Systems
Communication University of China (CN)
Remote-Sensing Image Classification
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon