97 / 2021-07-21 23:00:52

Method for Denoising Hyperspectral Images Based on Low Rank Theory-Sparse Representation

Special Session > SS7 - Data-driven Engineering Practices in The Fields of Energy, Environment, and Biomedicine

Various methods for denoising noisy hyperspectral images have been proposed, but the performance on key indicators shows that there is still room for improvement. Here, we propose a new denoising method for noisy hyperspectral images based on low-rank theory-sparse representation: standard deviation of the noise is estimated through singular value decomposition (SVD) on spatial-spectral joint matrix, and the low-noise matrix is extracted through principal component analysis (PCA) for spectral information decorrelation and information fusion. The low-noise data are sparsely filtered through the KSVD algorithm. Experimental results show that our model denoises excellently on hyperspectral images containing noise with different standard deviations, and the PSNR indicators are 5-10 dB ahead of the compared algorithms, showing great potential to optimize the expression of hyperspectral data.