STDF: Joint Spatiotemporal Differences Based on xLSTM Dendritic Fusion Network for Remote Sensing Change Detection

Change detection (CD) techniques aim to identify regional changes by analyzing spatiotemporal differences in remote sensing images. However, detecting subtle changes and addressing interference from complex background noise remain significant challenges. To address these issues, we propose a novel r...

Full description

Saved in:
Bibliographic Details
Main Authors: Yu Zhou, Shengning Zhou, Dapeng Cheng, Jinjiang Li, Zhen Hua
Format: Article
Language:English
Published: IEEE 2025-01-01
Series:IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
Subjects:
Online Access:https://ieeexplore.ieee.org/document/10937754/
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Change detection (CD) techniques aim to identify regional changes by analyzing spatiotemporal differences in remote sensing images. However, detecting subtle changes and addressing interference from complex background noise remain significant challenges. To address these issues, we propose a novel remote sensing CD network, STDF-CD, which integrates spatiotemporal difference modeling and dynamic feature fusion mechanisms to enhance performance in detecting subtle changes and improve robustness in complex background scenarios. Specifically, we design the STxLSTM, based on a state space model, to perform pixelwise scanning and capture spatiotemporal difference features from bi-temporal images, enabling the precise detection of subtle changes hidden within complex environments. In addition, we introduce a hierarchical dendritic coordination module based on the dendritic neuron model, which leverages a dynamic weighting mechanism to flexibly integrate multiscale feature maps. This approach not only mitigates the noise interference caused by traditional rigid fusion methods but also strengthens the separation of features between change regions and the background, thereby improving detection accuracy and model stability. Experiments show that STDF-CD achieves 91.92%, 94.90%, 87.93%, 81.53% F1 and 85.05%, 90.29%, 78.46%, 68.83% IoU on LEVIR-CD, WHU-CD, GZ-CD, and SYSU-CD datasets, respectively, outperforming existing methods, especially in detecting subtle changes and alleviating complex background noise.
ISSN:1939-1404
2151-1535