Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

RSDehamba: Lightweight Vision Mamba for Remote Sensing Satellite Image Dehazing

Huiling Zhou, Xianhao Wu, Hongming Chen, Xiang Chen, Xin He

TL;DR

This paper addresses remote sensing image dehazing (RSID) by introducing RSDehamba, a lightweight Mamba-based architecture that integrates a State Space Model (SSM) into a U‑Net backbone via Vision Dehamba Blocks (VDB) and enhances spatial awareness with a Direction-aware Scan Module (DSM). The key contributions are: (i) the first lightweight Mamba-based RSID network; (ii) the Vision Dehamba Block that enriches global context through SSM; (iii) the Direction-aware Scan Module that captures spatial haze distribution from multiple directions; and (iv) comprehensive ablation studies validating the effectiveness of SSM and DSM. Experiments on the SateHaze1K dataset demonstrate state-of-the-art performance with about 1.80M parameters and low FLOPs, indicating strong efficiency for resource-constrained RS applications. The work advances RSID by enabling accurate haze removal with reduced computational burden, potentially enabling real-time or onboard processing for remote sensing platforms.

Abstract

Remote sensing image dehazing (RSID) aims to remove nonuniform and physically irregular haze factors for high-quality image restoration. The emergence of CNNs and Transformers has taken extraordinary strides in the RSID arena. However, these methods often struggle to demonstrate the balance of adequate long-range dependency modeling and maintaining computational efficiency. To this end, we propose the first lightweight network on the mamba-based model called RSDhamba in the field of RSID. Greatly inspired by the recent rise of Selective State Space Model (SSM) for its superior performance in modeling linear complexity and remote dependencies, our designed RSDehamba integrates the SSM framework into the U-Net architecture. Specifically, we propose the Vision Dehamba Block (VDB) as the core component of the overall network, which utilizes the linear complexity of SSM to achieve the capability of global context encoding. Simultaneously, the Direction-aware Scan Module (DSM) is designed to dynamically aggregate feature exchanges over different directional domains to effectively enhance the flexibility of sensing the spatially varying distribution of haze. In this way, our RSDhamba fully demonstrates the superiority of spatial distance capture dependencies and channel information exchange for better extraction of haze features. Extensive experimental results on widely used benchmarks validate the surpassing performance of our RSDehamba against existing state-of-the-art methods.

RSDehamba: Lightweight Vision Mamba for Remote Sensing Satellite Image Dehazing

TL;DR

This paper addresses remote sensing image dehazing (RSID) by introducing RSDehamba, a lightweight Mamba-based architecture that integrates a State Space Model (SSM) into a U‑Net backbone via Vision Dehamba Blocks (VDB) and enhances spatial awareness with a Direction-aware Scan Module (DSM). The key contributions are: (i) the first lightweight Mamba-based RSID network; (ii) the Vision Dehamba Block that enriches global context through SSM; (iii) the Direction-aware Scan Module that captures spatial haze distribution from multiple directions; and (iv) comprehensive ablation studies validating the effectiveness of SSM and DSM. Experiments on the SateHaze1K dataset demonstrate state-of-the-art performance with about 1.80M parameters and low FLOPs, indicating strong efficiency for resource-constrained RS applications. The work advances RSID by enabling accurate haze removal with reduced computational burden, potentially enabling real-time or onboard processing for remote sensing platforms.

Abstract

Remote sensing image dehazing (RSID) aims to remove nonuniform and physically irregular haze factors for high-quality image restoration. The emergence of CNNs and Transformers has taken extraordinary strides in the RSID arena. However, these methods often struggle to demonstrate the balance of adequate long-range dependency modeling and maintaining computational efficiency. To this end, we propose the first lightweight network on the mamba-based model called RSDhamba in the field of RSID. Greatly inspired by the recent rise of Selective State Space Model (SSM) for its superior performance in modeling linear complexity and remote dependencies, our designed RSDehamba integrates the SSM framework into the U-Net architecture. Specifically, we propose the Vision Dehamba Block (VDB) as the core component of the overall network, which utilizes the linear complexity of SSM to achieve the capability of global context encoding. Simultaneously, the Direction-aware Scan Module (DSM) is designed to dynamically aggregate feature exchanges over different directional domains to effectively enhance the flexibility of sensing the spatially varying distribution of haze. In this way, our RSDhamba fully demonstrates the superiority of spatial distance capture dependencies and channel information exchange for better extraction of haze features. Extensive experimental results on widely used benchmarks validate the surpassing performance of our RSDehamba against existing state-of-the-art methods.
Paper Structure (12 sections, 9 equations, 4 figures, 3 tables)

This paper contains 12 sections, 9 equations, 4 figures, 3 tables.

Table of Contents

  1. Introduction
  2. PROPOSED METHOD
  3. Network Architecture
  4. Vision Dehamba Block
  5. State Space Model
  6. Direction-aware Scan Module
  7. EXPERIMENT RESULTS
  8. Implementation Details
  9. Qualitative Comparison
  10. Quantitative Comparison
  11. Ablation Study
  12. Conclusion

Figures (4)

  • Figure 1: Model complexity and performance comparisons of the proposed method and other state-of-the-art models on the Sate1K dataset in terms of PSNR, model parameters and FLOPs. The area of each circle denotes the number of FLOPs. Here, FLOP calculation is based on image sizes of $256 \times 256$. Our RSDhamba achieves the SOTA performance.
  • Figure 2: Overall architecture of RSDhamba. (a) RSDhamba consists of multiple Vision Dehamba Blocks (VDBs) using the U-Net architecture for image defogging tasks. (b) The VDB mainly consists of the State Space Model (SSM): a module for capturing local and global features. (c) An important component of SSM: the Direction-aware Scan Module (DSM) performs perceptual modeling by scanning from four directions.
  • Figure 3: The scanning mechanism of DSM is the fusion of regional haze features from four different directions to better perceive the distribution of spatial changes in haze.
  • Figure 4: Comparison of visual results for the SateHaze1k dataset, with colored boxes highlighting obvious differences. Zoom in for the best view.