Abstract: The image noise distribution of each band in a hyperspectral image is complex, and it is difficult for the traditional denoising method to achieve the desired effect. To address this problem, a new hyperspectral denoising method is proposed, based on the selection of the band combined with elite atomic joint dictionary learning. Firstly, the original hyperspectral data is reduced by band selection while retaining the main physical information of the spectrum. Then, the K-SVD dictionary learning is performed on each band of the selected image. Finally, the dictionary of each band learning is selected by the elite atom. This strategy generates a joint dictionary, proposes a dictionary learning denoising algorithm with adaptive dictionary length characteristics, and applies it to hyperspectral noisy images for denoising processing. Experiments on hyperspectral remote sensing images show that the peak signal-to-noise ratio (PSNR) of the image after denoising is improved compared with CFS, CFS-SRNS, and CFS-KSVD.

Key words: dictionary learning, sparse representation, hyperspectral image, image de-noising