研究目的
To develop an effective denoising algorithm for hyperspectral images using dictionary learning and sparse coding in the spectral domain to reduce noise while preserving spatial and spectral details.
研究成果
The HyDeSpDLS algorithm effectively denoises hyperspectral images by leveraging spectral dictionary learning and sparse coding with total variation regularization, outperforming existing methods in preserving spatial and spectral details, though it requires optimization for faster execution.
研究不足
The proposed algorithm has higher computational time compared to low-rank-based methods like LRMR and NAILRMA, and future work is needed to improve its speed.
1：Experimental Design and Method Selection:
The study uses an online spectral dictionary learning method (OSDL) to train a dictionary from noisy HSI data, employing pixel spectral vectors as training data. Sparse coding is performed using SpaRSAL-TV, which incorporates total variation regularization to utilize spatial-contextual information.
2：Sample Selection and Data Sources:
Synthetic data from the Washington DC Mall scene (256x256 pixels, 191 bands) and real-world Indian Pine data (145x145 pixels, 220 bands) are used. Noise types include Gaussian i.i.d, Gaussian non-i.i.d, and Poissonian noise.
3：List of Experimental Equipment and Materials:
A laptop with Intel Core i7-7700HQ CPU and 16 GB RAM running MATLAB R2014a is used for implementation.
4：Experimental Procedures and Operational Workflow:
The noisy HSI is expanded into a pixel spectral matrix, a spectral dictionary is trained using OSDL, sparse codes are computed with SpaRSAL-TV, and the denoised image is reconstructed. Performance is evaluated using PSNR and SSIM indices.
5：Data Analysis Methods:
Quantitative analysis involves calculating mean PSNR and SSIM values, and qualitative assessment through visual inspection of denoised images and spectral signatures.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
