论文部分内容阅读
目的高光谱遥感图像常存在多种不同程度的退化,进而影响到后续的应用,因此,对高光谱图像进行噪声水平估计具有重要意义。在实际情况中,不同波段的图像噪声水平常有所差异,需要针对不同谱通道的特性差异进行噪声估计。因此,本文提出一种基于低秩表达的噪声水平估计算法。方法该算法首先利用多波段图像间的光谱相关性,建立高光谱数据的低秩表达模型;再通过该模型对各波段的噪声及其水平进行估计,并根据需要检测并剔除被噪声淹没的无效波段。结果在多组高光谱数据上进行模拟和真实实验,证明本文算法能够准确估计高光谱图像的谱通道噪声水平。结论本文算法挖掘了低秩表达在高光谱应用中的特性,在利用波段间相关性进行全局处理的同时,也能保留波段间的差异,具有较强的鲁棒性;在合适的阈值范围内,无效波段的漏检率低至0,准确率高于80%。
Purpose Hyperspectral remote sensing images often have a variety of degrees of degradation, which will affect subsequent applications. Therefore, it is of great significance to estimate the noise level of hyperspectral images. In practice, the image noise levels of different bands often differ, and noise estimation needs to be performed according to the characteristics of different spectral channels. Therefore, this paper presents a noise level estimation algorithm based on low-rank representation. Methods The algorithm firstly uses the spectral correlation between multi-band images to establish a low-rank representation model of hyperspectral data. Then the noise and its level in each band are estimated by this model. Then, the algorithm is used to detect and reject the noise submerged Band. Results Simulation and real experiments on multiple sets of hyperspectral data show that the proposed algorithm can accurately estimate the spectral channel noise level of hyperspectral images. Conclusions The algorithm of this paper finds out the characteristics of low rank representation in hyperspectral applications. While utilizing the correlation between bands for global processing, it can also preserve the differences between bands and has strong robustness. Within the appropriate threshold range , The invalid detection rate of the band as low as 0, the accuracy rate is higher than 80%.