基于稀疏表示的谱线自动提取方法

A Method for Auto-Extraction of Spectral Lines Based on Sparse Representation

谱线提取在光谱分析中起着非常重要的作用,它对后续的光谱分类和参数测量有着直接的影响。文章提出了一种基于稀疏表示的谱线自动提取方法。首先,用基于稀疏表示的小波去噪方法去除噪声,该方法通过对光谱信号对应的小波系数进行稀疏化处理来达到去噪的目的,其优点是在处理小波系数时虽然将其作为整体进行考虑,但依然能保持小波系数的局部特性不变,所以在去噪的同时很好地保持了特征谱线的信息。其次,利用小波变换与样条拟合相结合的方法拟合出较为满意的伪连续谱,该方法在拟合过程中,先将强谱线扣除掉,从而使得拟合结果非常接近真实的连续谱。最后,通过对归一化后的谱线光谱设置自适应局部阈值来提取特征谱线。实验结果表明该方法切实有效。

A new method for auto--extraction of spectral lines based on sparse representation is presented in the present paper. Firstly, the authors proposed a wavelet denoising scheme using a new theory called sparse representation for noise removal. After performing wavelet transform on the spectral signal, this method implements noise removal by solving an Optimization problem, which makes the wavelet coefficients at each scale sparsest. The proposed method not only takes the structure properties in the wavelet coefficients into consideration, but also can well maintain the local characteristics of wavelet coefficients. Therefore it can effectively keep the information of featured spectral lines during the process of denoising. Secondly, the authors got satisfying continua by respectively utilizing the wavelet transform method and spline fitting method. The strong spectral lines were firstly removed from the given spectrum with wavelet transform, leading to the result that the obtained continuum approximated the real one very well. Finally, the spectrum was divided point to point by the obtainable continuum and the normalized spectrum was obtained. And then spectral lines were extracted from the normalized spectrum by using adaptive local thresholding scheme. Experimental results show that the proposed method is effective and efficient in the application of auto-extraction of spectral lines. The authors＇ method will be also helpful for the automatic classification of astronomical spectra.