PF-Trans: Physics-Embedded Frequency-Aware Transformer for Spectral Reconstruction

2026-06-08

AI Summary

This study addresses the severe spectral aliasing in snapshot broadband filter array imaging caused by complex mask modulation, a challenge that existing deep learning methods struggle to model due to their limited capacity to capture global frequency-domain degradation. To overcome this limitation, the authors propose a dual-domain Transformer architecture that integrates physical priors with frequency-aware modeling. The framework explicitly embeds the imaging physics through a mask injection mechanism and a grayscale consistency loss, while a parallel FFT branch jointly processes spatial and spectral information. This design enables, for the first time within a Transformer, physically driven suppression of frequency-domain aliasing. Evaluated on multiple remote sensing datasets, the method significantly outperforms state-of-the-art approaches, achieving a PSNR of 48.50 dB on the GF-5 Shanghai dataset while maintaining high reconstruction fidelity and physical consistency.

Abstract

Snapshot Broadband Filter Array (BFA) imaging provides high light throughput for spectral reconstruction but introduces severe spectral aliasing due to complex modulation. Current deep learning approaches, limited to spatial denoising, often fail to address the global frequency-specific degradations caused by the mask structure. To address this, we propose a Physics-embedded Frequency-aware Transformer (PF-Trans) for high-fidelity remote sensing spectral reconstruction. Our method explicitly integrates the physical sensing model through mask injection and a gray-scale consistency loss to ensure physical fidelity. Furthermore, we introduce a Dual-domain Block with a parallel Fast Fourier Transform (FFT) branch, enabling the network to perceive and suppress aliasing artifacts in the frequency domain. Extensive experiments on multiple datasets demonstrate that PF-Trans achieves state-of-the-art performance, achieving a Peak Signal-to-Noise Ratio (PSNR) of up to 48.50 dB on the GF-5 Shanghai dataset, significantly outperforming comparison methods.

Problem

• spectral reconstruction
• spectral aliasing
• broadband filter array
• frequency-specific degradation
• remote sensing

Innovation

• Physics-embedded
• Frequency-aware
• Spectral Reconstruction
• Transformer
• Aliasing Suppression