Towards UAV Image Dehazing: A UAV Atmospheric Scattering Model, Benchmark, and Geometry-Aware Deep Unfolding Network

2026-06-15 · Source: Computer Vision and Pattern Recognition · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems · Depth: Expert, quick

Summary

A new research initiative introduces a UAV Atmospheric Scattering Model (UASM) to tackle significant challenges in dehazing UAV imagery. These include modeling spatially non-uniform haze and the unavailability of paired real-world hazy/clean images. UASM explicitly integrates flight altitude, viewing pitch, and extinction to characterize non-uniform haze distribution. Based on UASM, the Geometry-aware Proximal Deep Unfolding Network (GP-DUN) was developed. GP-DUN comprises a Latent Geometry Estimator, a Geometry-aware Gradient Descent Module embedding UASM, and a Pooling-Expert Proximal Mapping Module. This module restores textures and structures beyond explicit physical modeling. Additionally, UASM-HazeSet was created, providing controllable paired synthetic data and 2,285 real UAV haze images for testing. Experiments confirm GP-DUN's superior performance against existing methods on both synthetic and real UAV haze benchmarks.

Key takeaway

If you are a Computer Vision Engineer developing robust UAV imaging systems, integrate the UAV Atmospheric Scattering Model (UASM). Also, implement the Geometry-aware Proximal Deep Unfolding Network (GP-DUN). This combination directly addresses non-uniform haze in UAV imagery. It offers superior dehazing performance. You should also utilize the UASM-HazeSet for generating realistic synthetic training data. This enhances your model's generalization capabilities in real-world hazy conditions, improving overall system reliability.

Key insights

A new UAV atmospheric scattering model and deep unfolding network effectively dehaze UAV images by accounting for non-uniform haze geometry.

Principles

• Haze distribution is non-uniform in UAV imaging.
• Incorporate flight geometry for accurate haze modeling.
• Physics-driven networks improve dehazing performance.

Method

GP-DUN uses a Latent Geometry Estimator, a Geometry-aware Gradient Descent Module embedding UASM, and a Pooling-Expert Proximal Mapping Module to restore textures.

In practice

• Use UASM to model non-uniform UAV haze.
• Apply GP-DUN for robust UAV image dehazing.
• Utilize UASM-HazeSet for synthetic data generation.

Topics

• UAV Imaging
• Image Dehazing
• Atmospheric Scattering Model
• Deep Unfolding Networks
• Computer Vision
• Synthetic Data Generation

Best for: AI Scientist, Computer Vision Engineer, Research Scientist

Related on AIssential

• Dehaze-GaussianImage: Zero-Shot Dehazing via Efficient 2D Gaussian Splatting Representation — Dehaze-GaussianImage uses 2D Gaussian Splatting for zero-shot dehazing, modeling haze as evolvable anisotropic Gaussian fields.
• HDST-GNN: Heterogeneous Dynamic Spatiotemporal Graph Neural Networks for Multi-Object Tracking in UAV Aerial Imagery — HDST-GNN improves multi-object tracking in UAV imagery by modeling heterogeneous object states and dynamic spatial context.
• Multi-Branch Non-Homogeneous Image Dehazing via Concentration Partitioning and Image Fusion — Decomposing non-homogeneous haze into homogeneous sub-problems enables targeted, multi-branch image enhancement and fusion.
• ErA: Error-Aware Deep Unrolling Network for Single Image Defocus Deblurring — ErA employs an error-aware deep unrolling network for state-of-the-art single-image defocus deblurring.
• NTIRE 2026: A Multi-Track Study in Image Restoration — NTIRE workshops benchmark image restoration models against real-world degradations using rigorous evaluation protocols.
• Leveraging Large Vision Model for Multi-UAV Co-perception in Low-Altitude Wireless Networks — A novel framework optimizes multi-UAV perception by sparsifying visual data and using LVMs with DRL for efficient communication.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Computer Vision and Pattern Recognition.