[Paper Review] CDMamba: Incorporating Local Clues into Mamba for Remote Sensing Image Binary Change Detection

Haotian Zhang, Keyan Chen+4 more|arXiv (Cornell University)|Jun 6, 2024

Remote-Sensing Image Classification18 citations

TL;DR

CDMamba fuses global features from Mamba with local detail via SRCM and adaptive fusion to improve binary change detection on remote sensing data, achieving state-of-the-art results on multiple datasets.

ABSTRACT

Recently, the Mamba architecture based on state space models has demonstrated remarkable performance in a series of natural language processing tasks and has been rapidly applied to remote sensing change detection (CD) tasks. However, most methods enhance the global receptive field by directly modifying the scanning mode of Mamba, neglecting the crucial role that local information plays in dense prediction tasks (e.g., binary CD). In this article, we propose a model called CDMamba, which effectively combines global and local features for handling binary CD tasks. Specifically, the Scaled Residual ConvMamba (SRCM) block is proposed to utilize the ability of Mamba to extract global features and convolution to enhance the local details to alleviate the issue that current Mamba-based methods lack detailed clues and are difficult to achieve fine detection in dense prediction tasks. Furthermore, considering the characteristics of bi-temporal feature interaction required for CD, the Adaptive Global Local Guided Fusion (AGLGF) block is proposed to dynamically facilitate the bi-temporal interaction guided by other temporal global/local features. Our intuition is that more discriminative change features can be acquired with the guidance of other temporal features. Extensive experiments on five datasets demonstrate that our proposed CDMamba is comparable to the current methods (such as the F1/IoU scores are improved by 2.10%/3.00% and 2.44%/2.91% on LEVIR+CD and CLCD, respectively). Our code is open-sourced at https://github.com/zmoka-zht/CDMamba.

Motivation & Objective

• Motivate improved dense prediction in remote sensing change detection by leveraging both global context and local details.
• Propose a CDMamba architecture that combines scaled residual ConvMamba (SRCM) for local+global features with an adaptive global-local fusion for bi-temporal interaction.
• Show that integrating local clues into Mamba yields superior performance on standard CD benchmarks.
• Provide open-source code to enable reproducibility and further research.

Proposed method

• Introduce Scaled Residual ConvMamba (SRCM) to blend ConvNet-style local processing with Mamba's global modeling.
• Develop Adaptive Global Local Guided Fusion (AGLGF) to dynamically fuse bi-temporal features guided by global and local cues.
• Use a two-branch ConvMamba-based encoder with multi-scale features and a decoder to produce differential change representations.
• Employ absolute subtraction of multi-scale features to form differential features for decoding.
• Train with a joint cross-entropy and dice loss to optimize pixel-wise change detection.

Experimental results

Research questions

• RQ1 Can local information be effectively integrated into Mamba-based change detection to improve dense predictions?
• RQ2 Does adaptive fusion guided by other temporal features enhance bi-temporal interaction for CD tasks?
• RQ3 How does CDMamba perform compared with CNN-, Transformer-, and other Mamba-based CD methods on standard benchmarks?

Key findings

• CDMamba achieves state-of-the-art performance compared with CNN-, Transformer-, and Mamba-based methods across WHU-CD, LEVIR-CD, and LEVIR+-CD datasets.
• Compared to prior Mamba methods, CDMamba yields improvements in F1 and IoU, demonstrating a balanced and robust change detection capability.
• Qualitative analyses show sharper change edges and better performance on small changes and complex scenes compared with competing methods.