研究目的
To propose a new low-rank spectral nonlocal approach (LRSNL) for the simultaneous removal of a mixture of different types of noises in hyperspectral images (HSI), improving their visual quality and the accuracy in target detection or classification.
研究成果
The proposed LRSNL approach effectively removes mixed noises and preserves fine spatial structures in HSI. The experiments demonstrate its superiority over LR matrix recovery method (LRMR) and SNL without precleaning. The importance of the precleaning step is highlighted for better clustering and restoration.
研究不足
LRSNL treats all spectral bands the same and uses the average of all bands to calculate similarities between patches, which may not be optimal when spectral bands are of different importance.
1：Experimental Design and Method Selection:
The LRSNL approach combines the low-rank (LR) property of HSI with a spectral nonlocal (SNL) method for restoration. The LR property is used for precleaning patches, and SNL is applied for final restoration, considering both spectral and spatial information.
2：Sample Selection and Data Sources:
Synthetic and real HSI datasets are used, including an Indian Pine dataset and an EO-1 Hyperion image dataset.
3：List of Experimental Equipment and Materials:
Not explicitly mentioned in the paper.
4：Experimental Procedures and Operational Workflow:
The HSI cube is divided into small patches, precleaned using the LR property, and then restored using the SNL method with weights calculated based on spectral and spatial information.
5：Data Analysis Methods:
Performance is evaluated using improved signal to noise ratio (ISNR) for each spectral band and visual comparison of restored 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
