Physics-Inspired Frequency-Decoupled Network for Remote Sensing Image Dehazing.

Area of Science:

• Remote Sensing
• Computer Vision
• Image Processing

Background:

• Remote sensing (RS) imagery is degraded by atmospheric scattering, causing loss of contrast and detail.
• Existing State Space Models (SSMs) for image restoration struggle with spectral-spatial coupling and lack physical constraints, leading to artifacts.

Purpose of the Study:

• To develop a physics-inspired network for high-fidelity remote sensing image dehazing.
• To address limitations of current SSMs in spectral-spatial coupling and physical interpretability.

Main Methods:

• Proposed PhysWave-SSN, a Physics-Inspired Frequency-Decoupled Network.
• Introduced Frequency-Aware Selection Gate (FASG) for adaptive frequency component recalibration.
• Developed Physics-Informed SSM (PI-SSM) with dynamic step size modulation based on haze density.
• Incorporated Luminance-Adaptive Fusion Module (LAFM) for spectral consistency.

Main Results:

• Achieved superior performance in RS image dehazing across multiple datasets.
• Demonstrated significant improvements in Peak Signal-to-Noise Ratio (PSNR), with a maximum gain of 2.49 dB.
• Successfully preserved structural details and spectral fidelity in dehazed images.

Conclusions:

• PhysWave-SSN effectively overcomes limitations of existing methods for RS image dehazing.
• The physics-informed approach enhances model interpretability and performance.
• The proposed network ensures high-fidelity restoration of remote sensing imagery.