Single-Frame Blind-Pixel Detection and Mask-Constrained Correction of Spatial Heterodyne Interferograms Using Robust Principal Component Analysis

A spatial heterodyne spectrometer records two-dimensional interferograms using an area-array detector and reconstructs high-resolution spectra through Fourier transformation. Blind pixels caused by detector defects can disrupt interferometric structures and increase spectral reconstruction errors. To detect and correct blind pixels in a single interferogram without prior knowledge of their locations or spatial distributions, we propose a mask-constrained recovery method based on robust principal component analysis (MC-RPCA). The method separates low-rank interferometric information from sparse anomalous responses, constructs a blind-pixel mask using a median-absolute-deviation-based adaptive threshold, and replaces the detected anomalous responses with the corresponding low-rank estimates. The method was evaluated through single-type and mixed-type blind-pixel experiments using ideal interferograms, together with experiments using measured potassium- and xenon-lamp interferograms. In the ideal single-type blind-pixel experiments, the root-mean-square error of the corrected single-row reconstructed spectra was reduced to values ranging from 2.9923 × 10−9 to 2.2175 × 10−8. In the ideal mixed blind-pixel experiment, the precision, recall and F1-score all reached 1.0000. In the blind-pixel injection experiment using a measured xenon-lamp interferogram, the average spectral RMSE decreased from 0.18937 to 0.05367. The results demonstrate that MC-RPCA enables training-free detection and correction of mixed point and block blind pixels while mitigating their effects on interferometric structures and reconstructed spectra.