物理驱动的中波红外光谱压缩编码与重建

Key Points

• This study aims to develop a network architecture for high-accuracy reconstruction of mid-wave infrared gas absorption spectra.
• Developed a dual-branch network focusing on smooth background reconstruction and sharp absorption feature reconstruction.
• Utilized information fusion and physical quantity transformation for improved accuracy.
• Conducted experiments on 45 channels of gas absorption spectra within the 3.7~4.8 μm range.
• Achieved a peak signal-to-noise ratio greater than 28.159 dB.
• Spectral angle mapping improved to better than 0.053 rad.
• Reconstruction time for a 320×256 data cube was approximately 0.65 seconds.