SPNet: a size-variant progressive network for aero-optical thermal radiation effects correction

Area of Science:

• Aerospace Engineering
• Optical Physics
• Computer Vision

Background:

• High-speed aircraft airflow compression causes aero-optical thermal radiation, degrading optical cover image quality.
• Existing methods struggle with the complex and variable nature of these thermal radiation effects.

Purpose of the Study:

• To propose a novel network, SPNet, for correcting image quality degraded by aero-optical thermal radiation effects.
• To leverage the size-invariant characteristic of thermal radiation bias fields for effective correction.

Main Methods:

• Developed a size-variant progressive network (SPNet) with two sub-networks for coarse and accurate bias field correction.
• Introduced a multi-scale feature upsampling module (MFUM) for enhanced feature interaction.
• Proposed an adaptive feature fusion module (AFFM) for dynamic scale-based feature merging.
• Implemented a multi-head self-attention feature extraction module (MSFEM) for global information capture.

Main Results:

• SPNet progressively corrects degraded images by learning thermal radiation bias fields.
• The MFUM and AFFM modules effectively utilize multi-scale and inter-channel features.
• MSFEM extracts comprehensive global features, improving correction accuracy.
• Experimental results on simulated and real data show superior performance compared to state-of-the-art methods.

Conclusions:

• SPNet offers a robust solution for correcting aero-optical thermal radiation effects in aircraft imagery.
• The proposed modules enhance the network's ability to handle complex image degradations.
• The method demonstrates significant improvements in image quality restoration for high-speed platforms.