研究目的
To compare least square (LS) weighted regularization in spectral domain with spatial least square and total variation (TV) denoising techniques for hyperspectral image denoising.
研究成果
Spectral least square denoising is superior to spatial LS and TV denoising in preserving image contrast and edges, with higher SNR and SSIM values. Denoising in the spectral domain is more efficient, and grouping bands based on noise levels can optimize the process. Future work should integrate denoised bands into classification and unmixing tasks to improve accuracy.
研究不足
The control parameter λ is fixed based on spectral LS method, which may not be optimal for spatial LS and TV methods, potentially leading to suboptimal performance in those cases. The study uses specific datasets and noise types (Gaussian), which may limit generalizability to other types of noise or images. Computational efficiency is improved by denoising only noisy bands, but the IBBC method's accuracy in band selection could be a constraint. Visual interpretation is subjective and may vary among observers.
1：Experimental Design and Method Selection:
The study compares three denoising methods: spectral least square (LS), spatial LS, and total variation (TV) denoising. The methods are applied to hyperspectral image datasets, with noise simulated in some cases. The Inter Band Blockwise Correlation (IBBC) method is used to automatically detect noisy spectral bands, reducing computational time by denoising only noisy bands. Control parameter λ is varied to optimize denoising, with evaluation based on visual interpretation, Signal-to-Noise Ratio (SNR), Peak-Signal-to-Noise Ratio (PSNR), and Structural Similarity (SSIM) Index.
2：Sample Selection and Data Sources:
Four standard hyperspectral datasets are used: Washington DC Mall, Indian Pines, Salinas-A scene, and University of Pavia. These datasets were captured by sensors such as AVIRIS and ROSIS, with varying numbers of spectral bands and spatial resolutions. Noise is simulated in the Washington DC Mall dataset by adding Gaussian noise with constant mean and varied variance.
3：List of Experimental Equipment and Materials:
Hyperspectral image datasets (Washington DC Mall, Indian Pines, Salinas-A, University of Pavia), sensors (AVIRIS for Indian Pines and Salinas-A, ROSIS for University of Pavia), and computational tools for implementing denoising algorithms and metrics calculation.
4：Experimental Procedures and Operational Workflow:
The HSI is vectorized into a 2D matrix. Denoising is applied column-wise and then row-wise for LS methods. Noisy bands are identified using IBBC and thresholding. Denoising is performed with varying λ values, and results are compared using visual inspection, SNR, PSNR, and SSIM. The workflow includes noise simulation, band selection, denoising application, and quality assessment.
5：Data Analysis Methods:
SNR and PSNR are calculated to assess denoising quality mathematically. SSIM is used for visual quality assessment based on human visual system parameters. Results are analyzed through plots (e.g., PSNR vs. λ) and tables comparing metrics across methods and datasets.
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
