高光谱图像(HSI)中的噪声去除是遥感技术中的一项基础而关键的任务,它对于图像的后续处理和分析至关重要。本项研究针对高光谱图像的去噪挑战,针对张量纤维秩约束优化与即插即用正则化的去噪技术对其中的不足进行了改进,即根据条带噪声...高光谱图像(HSI)中的噪声去除是遥感技术中的一项基础而关键的任务,它对于图像的后续处理和分析至关重要。本项研究针对高光谱图像的去噪挑战,针对张量纤维秩约束优化与即插即用正则化的去噪技术对其中的不足进行了改进,即根据条带噪声的组稀疏性质,通过L2-1范数对噪声中条带噪声组稀疏性质进行描述。有效提升了以往L1范数刻画条带噪声的去噪能力。最后通过应用乘子交替方向法(ADMM)来解决这一非凸优化问题。在多个遥感图像数据集上进行的实验验证了该方法在峰值信噪比(PSNR)和结构相似度(SSIM)等评价标准上的优越性,证明了其在处理复杂噪声条件下的高效性和广泛的应用前景。The noise removal in hyperspectral images (HSI) is a fundamental and crucial task in remote sensing technology, which is crucial for the subsequent processing and analysis of images. This study addresses the denoising challenge of hyperspectral images by improving the denoising techniques of tensor fiber rank constrained optimization and plug and play regularization. Based on the sparsity of band noise, the L2-1 norm is used to describe the band noise in the noise. It has improved the denoising ability of previous L1 norm characterization of stripe noise. Finally, by applying the Multiplier Alternating Directions Method (ADMM) to solve this non convex optimization problem, this method achieved a significant improvement in computational efficiency. Experiments conducted on multiple remote sensing image datasets have verified the superiority of this method in evaluation criteria such as peak signal-to-noise ratio (PSNR) and structural similarity (SSIM), demonstrating its efficiency and broad application prospects in dealing with complex noise conditions.展开更多
文摘高光谱图像(HSI)中的噪声去除是遥感技术中的一项基础而关键的任务,它对于图像的后续处理和分析至关重要。本项研究针对高光谱图像的去噪挑战,针对张量纤维秩约束优化与即插即用正则化的去噪技术对其中的不足进行了改进,即根据条带噪声的组稀疏性质,通过L2-1范数对噪声中条带噪声组稀疏性质进行描述。有效提升了以往L1范数刻画条带噪声的去噪能力。最后通过应用乘子交替方向法(ADMM)来解决这一非凸优化问题。在多个遥感图像数据集上进行的实验验证了该方法在峰值信噪比(PSNR)和结构相似度(SSIM)等评价标准上的优越性,证明了其在处理复杂噪声条件下的高效性和广泛的应用前景。The noise removal in hyperspectral images (HSI) is a fundamental and crucial task in remote sensing technology, which is crucial for the subsequent processing and analysis of images. This study addresses the denoising challenge of hyperspectral images by improving the denoising techniques of tensor fiber rank constrained optimization and plug and play regularization. Based on the sparsity of band noise, the L2-1 norm is used to describe the band noise in the noise. It has improved the denoising ability of previous L1 norm characterization of stripe noise. Finally, by applying the Multiplier Alternating Directions Method (ADMM) to solve this non convex optimization problem, this method achieved a significant improvement in computational efficiency. Experiments conducted on multiple remote sensing image datasets have verified the superiority of this method in evaluation criteria such as peak signal-to-noise ratio (PSNR) and structural similarity (SSIM), demonstrating its efficiency and broad application prospects in dealing with complex noise conditions.
