Dual-prior network improves clearing haze in remote sensing images
DPSF-Net: A Dual-Prior Spatial-Frequency Network for Real-World Remote Sensing Image Dehazing
Computer Vision and Pattern RecognitionArtificial Intelligence
Summary
Hazy conditions in satellite and aerial images make it hard to see details clearly and can distort colors. The authors created a computer method called DPSF-Net that uses both the usual color images and special maps showing haze patterns to clean up these images better. Their approach combines features from the image and its frequency details to separate large hazy areas from important local details. Tests showed that DPSF-Net works very well on real-world hazy remote sensing images and keeps a good balance between quality and computing cost.
What this means in practice
- •For remote sensing engineers: Improve clarity and color accuracy of satellite and aerial images for environmental monitoring under hazy conditions using a dual-prior dehazing method.
- •For urban planning teams: Enhance visibility of ground features in remote sensing images affected by atmospheric haze to support better decision making in urban development.
Authors
Mei Lu, Shangliang Shao, Shanliang Yao
Abstract
Real-world remote sensing image dehazing (RSID) remains challenging because atmospheric scattering, spatially non-uniform haze and colour distortion jointly degrade structural and spectral information. Most deep learning methods rely on RGB inputs and spatial-domain feature extraction, which limits their ability to separate global background haze from local surface details. Here, we propose DPSF-Net, a dual-prior spatial-frequency network built on MCAF-Net for real-world RSID. The network uses hazy RGB images and dark channel prior (DCP) maps as joint inputs, allowing physical degradation cues to guide end-to-end feature learning. A spatial-frequency residual interaction block introduces a FourierUnit branch into multi-directional spatial interaction to model large-scale haze components. A prior-guided feature attention module adaptively fuses prior and attention features to reduce colour shift and structural distortion. A selective kernel complementary fusion module screens multi-scale skip features through bidirectional residual complementary gating and selective kernel fusion. Extensive experiments demonstrate that DPSF-Net achieves state-of-the-art performance on the real-world RRSHID remote sensing image dehazing benchmark and remains competitive across multiple synthetic datasets. Moreover, the proposed method strikes a favourable balance among restoration quality, parameter count and computational complexity, supporting the effectiveness of dual-prior spatial-frequency modelling.