.png){kind=icon}
.png){kind=icon}
Return
Temporal Difference Learning via Frequency–Spatial Interaction for Remote Sensing Change Detection
.png){kind=icon}
PRE
AI
X
刘
J
X
肖
DOI：10.1109/lgrs.2026.3712317.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Remote sensing change detection (RSCD) aims to identify surface variations from bitemporal images. However, temporal difference learning in complex remote sensing scenes remains challenging, since nonchange regions often exhibit feature discrepancies caused by illumination variation, seasonal shifts, and background clutter, which easily induce pseudochanges and blurry boundaries. To address this issue, we propose a temporal difference learning via frequency–spatial interaction (TDFSI) framework for RSCD. In particular, we first design a temporal difference enhancement module (TDEM) to strengthen change-sensitive discrepancy cues from bitemporal features while suppressing unstable background responses. Then, a frequency-domain change refinement (FCR) module is introduced to refine the learned temporal difference representations in the spectral domain, so as to enforce global consistency, reduce pseudochange noise, and sharpen object boundaries. Finally, the refined multilevel features are fed into a UPerNet-based change decoder to generate the final prediction. Extensive experiments on the LEVIR-CD and SYSU-CD datasets demonstrate the effectiveness of the proposed method, achieving an <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$F1$ </tex-math></inline-formula>-score of 92.13% and an intersection over union (IoU) of 84.70% on LEVIR-CD. These results validate that frequency–spatial interaction can effectively enhance temporal difference learning for robust and precise RSCD (Code available at: <uri xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">https://github.com/lxllzzy/TDFSI-RSCD</uri>).
Keywords:
Boundary refinement
frequency–spatial interaction
remote sensing change detection (RSCD)
temporal difference learning
Journal
I
IF:
4.4
Papers:
486
Citations:
0
Organization
Cited Papers