红外光和可见光图像的融合在视频监控、目标跟踪等方面发挥着越来越重要的作用。为了得到融合效果更好的图像，提出了一种新的基于鲁棒性低秩表示的图像分解与深度学习结合的方法。首先，利用鲁棒性主成分分析对训练集图像进行去噪处理，利用快速的潜在低秩表示学习提取突出特征的稀疏矩阵，并对源图像进行分解，重构形成低频图像和高频图像。然后，低频部分利用自适应加权策略进行融合，高频部分利用深度学习的 VGG-19网络进行融合。最后，将新的低频图像与新的高频图像进行线性叠加，得到最后的结果。实验验证了本文提出的图像融合算法在主观评价与客观评价上均具有一定的优势。Based on the Decomposition of Robust Principal Component Analysisand Latent Low Rank Representation

In this paper, the Raman spectrum signal de-noising based on stationary wavelet transform is discussed. Haar wavelet is selected to decompose the Raman spectrum signal for several levels based on stationary wavelet transform. The noise mean square {sigma}_j is estimated by the wavelet details at every level, and the wavelet details toward 0 by a threshold {sigma}_j (2\ln n)^{1/2} , where n is length of the detail, then recovery signal is reconstructed. Experimental results show this method not only suppresses noise effectively, but also preserves as many target characteristics of original signal as possible. This de-noising method offers a very attractive alternative to Raman spectrum signal noise suppress.