基于多尺度局部信噪比的拉曼谱峰识别算法  

Peak Detection Algorithm of Raman Spectra Based on Multi-Scale Local Signal-to-Noise Ratio

作　　者：姜承志[1,2] 孙强[1] 刘英[1] 梁静秋[3] 刘兵[1,2] 

机构地区：[1]中国科学院长春光学精密机械与物理研究所光电技术研发中心,吉林长春130033 [2]中国科学院大学,北京100049 [3]中国科学院长春光学精密机械与物理研究所应用光学国家重点实验室,吉林长春130033

出　　处：《光学学报》2014年第6期299-305,共7页Acta Optica Sinica

基　　金：国家自然科学基金(60977001);吉林省科技厅项目(20125092)

摘　　要：拉曼谱峰识别是拉曼光谱定性分析中的关键技术之一。针对现有方法的自动化程度不高、识别率低的问题,提出了一种新的基于多尺度局部信噪比(MLSNR)的拉曼谱峰识别算法。算法通过多尺度二阶差分运算,得到光谱的差分系数,再将差分系数除以估计出的噪声标准差,获得光谱的MLSNR矩阵,通过寻找MLSNR矩阵中的局部极大值形成的脊线来识别拉曼谱峰。算法采用自动阈值估计法去除噪声引起的局部极大值的干扰,可实现谱峰的自动化识别,不需设置任何参数。仿真实验结果表明:无论对单峰还是重叠峰,当拉曼谱峰信噪比大于等于6时MLSNR的谱峰识别准确率均高达100%,即使对处于检测限的单峰,仍有95%以上的识别准确率。MLSNR是一种切实可行的拉曼谱峰识别方法。Raman spectral peak recognition is one of the key technologies in qualitative analysis of Raman spectra. Due to the defects of low degree of automation and low recognition accuracy of the existing Raman spectral recognition methods, a new Raman peak recognition algorithm based on multi-scale local signal-to-noise ratio （MLSNR） is proposed. The algorithm gets the multi-scale second order difference coefficient of spectrum through multi-scale second order difference operation, then divides the multi-scale second order difference coefficient by the estimated noise standard deviation to obtain the MLSNR matrix of spectrum, and identifies Raman peaks by searching the ridges caused by local maxima in MLSNR matrix. The algorithm uses an automatic threshold estimation method to avoid the interference of local maximum caused by noise, and can recognize Raman peaks automatically without any parameter to be specified by human. The simulation result shows that no matter to singular peak or congested peaks, when the signal-to-noise ratio of Raman peak is greater than or equal to 6, the recognition accuracy of MLSNR algorithm is 100 %, even to the singular peak at the detection limit, the recognition accuracy is more than 95 %. MLSNR algorithm is a practical Raman spectral peak identification method.

关 键 词：光谱学 谱峰识别 多尺度局部信噪比 拉曼光谱 连续小波变换 

分 类 号：O433.4[机械工程—光学工程]