School of Information Science & Engineering, Lanzhou University, China.
Appl Spectrosc. 2018 Dec;72(12):1752-1763. doi: 10.1177/0003702818789695. Epub 2018 Aug 2.
Noise and fluorescent background are two major problems for acquiring Raman spectra from samples, which blur Raman spectra and make Raman detection or imaging difficult. In this paper, a novel algorithm based on wavelet transform that contains denoising and baseline correction is presented to automatically extract Raman signals. For the denoising section, the improved conventional-scale correlation denoising method is proposed. The baseline correction section, which is performed after denoising, basically consists of five aspects: (1) detection of the peak position; (2) approximate second derivative calculation based on continuous wavelet transform is performed using the Haar wavelet function to find peaks and background areas; (3) the threshold is estimated from the peak intensive area for identification of peaks; (4) correction of endpoints, spectral peaks, and peak position; and (5) determine the endpoints of the peak after subtracting the background. We tested this algorithm for simulated and experimental Raman spectra, and a satisfactory denoising effect and a good capability to correct background are observed. It is noteworthy that this algorithm requires few human interventions, which enables automatic denoising and background removal.
噪声和荧光背景是从样品中获取拉曼光谱的两个主要问题，它们会使拉曼光谱模糊，使拉曼检测或成象变得困难。在本文中，提出了一种基于小波变换的新算法，该算法包含去噪和基线校正，可自动提取拉曼信号。在去噪部分，提出了改进的常规尺度相关去噪方法。去噪后的基线校正部分基本包括五个方面：（1）检测峰位置；（2）使用 Haar 小波函数进行基于连续小波变换的近似二次导数计算，以找到峰和背景区域；（3）根据峰密集区估计阈值，以识别峰；（4）校正端点、谱峰和峰位置；以及（5）减去背景后确定峰的端点。我们对模拟和实验拉曼光谱进行了测试，观察到了令人满意的去噪效果和良好的背景校正能力。值得注意的是，该算法需要的人工干预较少，能够自动进行去噪和背景去除。
