DehazeMamba: SAR-guided Optical Remote Sensing Image Dehazing with Adaptive State Space Model

Abstract: Optical remote sensing image dehazing presents significant challenges due to its extensive spatial scale and highly non-uniform haze distribution, which traditional single-image dehazing methods struggle to address effectively. While Synthetic Aperture Radar (SAR) imagery offers inherently haze-free reference information for large-scale scenes, existing SAR-guided dehazing approaches face two critical limitations: the integration of SAR information often diminishes the quality of haze-free regions, and the instability of feature quality further exacerbates cross-modal domain shift. To overcome these challenges, we introduce DehazeMamba, a novel SAR-guided dehazing network built on a progressive haze decoupling fusion strategy. Our approach incorporates two key innovations: a Haze Perception and Decoupling Module (HPDM) that dynamically identifies haze-affected regions through optical-SAR difference analysis, and a Progressive Fusion Module (PFM) that mitigates domain shift through a two-stage fusion process based on feature quality assessment. To facilitate research in this domain, we present MRSHaze, a large-scale benchmark dataset comprising 8,000 pairs of temporally synchronized, precisely geo-registered SAR-optical images with high resolution and diverse haze conditions. Extensive experiments demonstrate that DehazeMamba significantly outperforms state-of-the-art methods, achieving a 0.73 dB improvement in PSNR and substantial enhancements in downstream tasks such as semantic segmentation. The dataset is available at https://github.com/mmic-lcl/Datasets-and-benchmark-code.

Overview of DehazeMamba: SAR-Guided Optical Remote Sensing Image Dehazing

The paper introduces DehazeMamba, a SAR-guided dehazing network for optical remote sensing images, addressing significant challenges in the field due to extensive spatial scale haze and non-uniform distribution. Current single-image dehazing methods have shown limitations in effectively dealing with these issues, where SAR imagery provides haze-free information crucial for large-scale scenes. However, existing methods integrating SAR information have suffered from quality degradation in haze-free regions and domain shift instability. To confront these challenges, DehazeMamba incorporates a progressive haze decoupling fusion strategy.

Key Innovations

The two primary innovations in DehazeMamba are:

1. Haze Perception and Decoupling Module (HPDM): This module dynamically identifies haze-affected regions through optical-SAR difference analysis. By focusing on the semantic discrepancies between modalities, HPDM efficiently decouples features in haze-affected areas.
2. Progressive Fusion Module (PFM): To address domain shift challenges, PFM utilizes a two-stage fusion process based on feature quality assessment, allowing selective integration of SAR information into optical images to improve dehazing results without degrading quality in haze-free regions.

These architectural components underscore the paper's contribution to SAR-guided dehazing, enabling more effective utilization of complementary SAR information.

Dataset Contribution

The authors introduce MRSHaze, a large-scale benchmark dataset comprising 8,000 pairs of SAR-optical images, synchronized temporally and geo-registered precisely. The dataset offers high-resolution imagery with diverse haze conditions, facilitating the evaluation of dehazing methodologies within complex environments.

Experimental Results

The experimental evaluation of DehazeMamba demonstrates notable advancements. The network achieves a 0.73 dB improvement in PSNR over state-of-the-art methods and enhances downstream tasks such as semantic segmentation notably. This outcome reflects DehazeMamba's effective SAR-information integration targeted at enhancing haze-affected optical image regions without compromising the integrity of surrounding non-hazy areas.

Implications and Future Directions

The theoretical implication of this research lies in advancing understanding of multimodal fusion in remote sensing applications, where adaptive techniques such as HPDM and PFM provide innovative solutions to complex optical-SAR integration challenges. Practically, the network's ability to produce high-quality dehazed images enhances various domains, including military, forestry, and agricultural remote sensing analyses.

Regarding future developments, additional research could focus on expanding the network's capabilities to handle multi-modal fusion more efficiently, assessing potential techniques for further computational cost reduction. Exploration into incorporating other sensor modalities beyond SAR could also enrich the framework's adaptive capabilities in diverse imaging environments.

By introducing effective strategies for multimodal data integration and improving image quality preservation, DehazeMamba contributes significantly to remote sensing image processing. The provision of MRSHaze further places empirical support for ongoing research in the field, maintaining the trajectory toward more robust and comprehensive data-driven dehazing solutions.