Denoising method for spatial heterodyne interferograms based on a convolutional encoder-decoder and transformer

The spatial heterodyne spectrometer may be affected by noise during detection, causing target spectral features to be obscured and preventing the acquisition of required information. Therefore, effective methods are needed to reduce the impact of noise on spatial heterodyne interferograms. For spatial heterodyne interferograms with the same type of detection targets and highly similar interference information in each row, a denoising method based on a convolutional encoder-decoder and transformer is proposed, and the method is trained and tested using real data from the GF-5 satellite. Experimental results show that the method can effectively denoise the interferogram and then improve spectral data quality, while demonstrating superior denoising performance compared to other algorithms. Finally, the trained model is used to denoise interferograms with real noise from GF-5 data, showing good denoising effectiveness. This work has study significance and application value in reducing the impact of noise on spatial heterodyne spectral information and achieving high-precision remote sensing detection of targets.