xLSTM Interaction Multilevel SSM-Assisted Decoding Network for Remote Sensing Image Change Detection

Remote sensing change detection (RSCD) plays a crucial role in postdisaster reconstruction, environmental monitoring, and urban planning. As a result, it has become a research hotspot in the field of remote sensing image processing in recent years. With the advancements of convolutional neural netwo...

Full description

Saved in:
Bibliographic Details
Main Authors: Chunpeng Wu, Shuli Cheng, Anyu Du, Liejun Wang, Wenbin Tang
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/11082571/
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Remote sensing change detection (RSCD) plays a crucial role in postdisaster reconstruction, environmental monitoring, and urban planning. As a result, it has become a research hotspot in the field of remote sensing image processing in recent years. With the advancements of convolutional neural networks (CNNs) and Transformers in deep learning, the accuracy of RSCD has significantly improved. This improvement is largely attributed to the local feature capture capability of CNNs and the long-range dependency modeling capability of Transformers. However, change detection (CD) tasks involve both temporal and spatial dimensions. Most current deep learning models have limited ability to model the temporal relationships between bitemporal features and lack effective handling of redundant features. To address these limitations, this article proposes an xLSTM interaction multilevel SSM-assisted decoding network (xLMSD-Net). It utilizes extended long short-term memory (xLSTM) to construct a temporal-aware feature interaction module, which adaptively learns the spatiotemporal relationships between bitemporal images, contrasting and enhancing features from both time phases. In addition, we combine gating mechanism and state space model to design a multichannel feature optimization mechanism (MCFOM) for multilevel decoding. The MCFOM suppresses redundant information level by level, enhancing feature representation during image reconstruction and improving the model’s robustness and learning capacity. Extensive experiments on the BCDD, LEVIR-CD, and CDD change detection datasets show that xLMSD-Net outperforms other state-of-the-art CD methods across multiple performance metrics.
ISSN:1939-1404
2151-1535